Subscribe now

The real question behind China’s bride-price debate is not how much marriage should cost. It is how much uncertainty young people are being asked to bear.

I read Justin Sun’s open letter to Hu Xijin twice. If I were answering on Hu’s behalf, I would not begin with whether “a man should behave decently,” or whether 30 million yuan is really that much money for a billionaire. Those are side arguments. Sun is asking a harder question: when a young Chinese person decides to marry, what exactly is he committing himself to? What responsibilities will he assume, what costs will he bear, and what degree of certainty can he reasonably expect in return?

That is a real question. The problem is that Sun has placed several different questions into the same account. Before answering him, we need to separate them.

The first point is simple: Chinese men are under no legal obligation to pay caili, or bride price. Sun writes, “A bride price has to be paid one way or another.” That may accurately describe his experience of Chinese social custom. It does not accurately describe Chinese marriage law. The Civil Code establishes freedom of marriage and equality between men and women, and prohibits demanding property through marriage. The Supreme People’s Court’s 2024 judicial interpretation on disputes involving betrothal gifts did not establish a legally acceptable price for marriage. It was meant to resolve disputes over payments that had already been made. And the 2026 No. 1 Central Document again called for continued efforts to curb excessively high bride prices in rural China.

The direction of policy, in other words, is not to answer the question, “How much should a man pay?” It is to allow a custom to exist without allowing that custom to harden into an admission fee for marriage. So when Sun asks whether bride price should be a fixed amount or a percentage of a man’s wealth, the answer is: neither.

The law does not say that a man with 10 million yuan should pay 300,000, while a man worth one billion should pay 30 million. When courts decide whether a bride price is excessively high, they look at the circumstances: local income levels, the financial position of the payer’s family, local customs, how long the couple lived together, whether they had children, and how the money was actually used. A couple who lived together for years and raised a child is not legally in the same position as a couple who broke up shortly after becoming engaged. In cases published by the Supreme People’s Court, some payments have been ordered returned in full, some only in part, and some not at all after long-term cohabitation and the birth of children.

There is a reason the law works this way. Marriage is not a sale. If it were, things would actually be easier. A car has a price. A house has a closing date. A contract can specify delivery, default, and damages. Marriage cannot. Love, companionship, pregnancy, childbirth, childcare, care for aging parents, housework, career sacrifices, illness, loyalty and decades of shared life cannot all be priced on the wedding day. Economists call this an incomplete contract. Ordinary language says it better: two people cannot settle forty years of accounts before they have even begun to live together.

That brings us to Sun’s second question: How much should a man have to pay for childbirth? Again, there is no price.

Pregnancy and childbirth impose physical burdens on women that men do not bear. That is a biological fact. In practice, women also frequently absorb more of the costs that come afterward: childcare, career interruption, lost income and unpaid care work. Men, meanwhile, have obligations of their own—to support, raise, educate and care for their children. Chinese law treats the raising and protection of children as a shared parental responsibility. What it does not do is treat reproduction as a service purchased by men from women.

That distinction matters. A modern society should not respond to the unequal burdens of reproduction by inventing a private price for childbirth. It should distribute those burdens more broadly. Some are borne jointly by the couple, some by the extended family, and some should be borne socially, through childcare, education, health care, parental leave, housing policy and other public services.

If nearly all of those costs are pushed back onto the individual family, men and women will eventually sit across the table and start adding up their losses. The man counts the apartment, the bride price, the wedding and the expectation that he provide. The woman counts pregnancy, childbirth, career interruption, housework and years of caregiving. By the time the arithmetic is finished, both sides believe they are being asked to make the worse bargain. That may be one of the deepest problems in Chinese marriage today: love has become increasingly private, but marriage has begun to look increasingly like a major capital investment.

Sun’s third question deserves an equally clear answer: Must a man satisfy every demand a woman makes before or after marriage? Of course not.

The basic principle of a modern marriage is not that a man must satisfy a woman. Nor is it that a woman must obey a man. It is that two adults both have the right to say no. Whether to buy a home can be negotiated. Who manages the money can be negotiated. Whether to live with either set of parents can be negotiated. How many children to have can be negotiated. The division of housework should be negotiated too.

But bodily autonomy and personal dignity belong to a different category. Consent cannot be purchased. Marriage itself must be voluntary. A person who does not want to marry has the right not to marry. A person who does not want children has the right to refuse parenthood. A person who does not consent to a particular intimate act does not acquire an obligation to perform it merely because the other party spent money. Payment cannot convert another person’s body into a debt.

What happens, then, if the marriage never takes place? Now we are dealing with property law. If money was genuinely transferred as a betrothal gift in contemplation of marriage and the marriage did not occur, the payer may seek its return through the courts. Even after legal marriage, a court may order some money returned when the couple lived together only briefly and the payment was exceptionally large. The opposite is also true. If the couple lived together for years, had children, or used the money for their common life, a court is unlikely to apply a crude rule of “I paid you X, therefore you owe me X.”

So when Sun asks whether a man can go to court if the woman leaves, the answer is straightforward: of course he can. There is nothing indecent about filing a lawsuit. There is nothing indecent about asking a court to decide whether 30 million yuan constituted bride price, a gift, or some other form of property transfer.

According to a public statement released by Sun’s lawyer on August 27, 2026, the relevant property dispute has been accepted by a court. The amount at issue exceeds 30 million yuan. The defendant has challenged jurisdiction, and the case has not yet proceeded to a determination on the merits. There is no final judgment. That means outsiders are in no position to decide which payments legally qualify as bride price, whether any of them must be returned, or in what amount. That is the court’s job.

The harder question comes afterward. Sun asks: Do I have the right to tell the public my side of the story? I chose to speak. Was that wrong?

My answer is that he unquestionably has the right to speak. But the right to speak is not the right to publish every detail of another person’s private life. That distinction has been almost completely lost in this controversy.

“I gave her 30 million yuan, I want it back, and I believe I was treated unfairly” is a public claim. “I have filed suit, and I believe those payments should legally be treated as bride price” is also a public claim. But if one tries to prove that claim by publishing intimate conduct, private correspondence, bodily information, family details or other material the other person never agreed to make public, the dispute changes character. A court adjudicates property. The internet adjudicates character. And unlike a court, it often does so without rules of evidence, proportionality or appeal.

China’s Civil Code separately protects privacy, including private spaces, private activities and private information. Those rights do not disappear because two people once loved each other. They do not vanish because the relationship ended badly. And they do not automatically yield because one side believes he has been wronged.

So if I were answering Sun for Hu Xijin, I would put the point somewhat differently from “a man should be decent.” I would say: Take the money dispute to court. Do not turn the internet into a tribunal for the relationship itself.

This principle is not gendered. A man who believes he has lost money may sue, report wrongdoing, seek preservation of assets, produce evidence and publicly state his legal position. A woman has the same rights. But secrets entrusted to someone during an intimate relationship should not automatically become ammunition when the relationship collapses.

Otherwise, intimacy itself becomes dangerous. What you tell someone you love today may become material used against you tomorrow. A private photograph may become a public exhibit. A conversation held in confidence may become a viral screenshot. A bedroom shared by two people may become entertainment for millions. If people begin to believe that every private relationship carries that risk, they will behave accordingly: they will disclose less, trust less, commit later, and sometimes decide not to commit at all.

Sun ends his letter by connecting the bride-price dispute to China’s demographic decline. On this point, he raises a question worth taking seriously. China recorded only 7.92 million births in 2025. In 1990, the national birth rate was 21.06 per thousand, which, given the population at the time, meant more than 20 million births. In 2024, only 6.106 million couples registered marriages, a decline of 20.5 percent from the year before. These numbers describe a profound transformation in Chinese marriage and family formation.

But they do not prove that the “missing” millions of births were driven away by bride price. Demography does not work that way. People are marrying later, marrying less and having fewer children for many reasons at once: longer education, urbanization, expensive housing, insecure employment, high childcare costs, greater professional opportunities for women, changing expectations of marriage, rising ages at first marriage and, increasingly, the simple fact that there are fewer young adults available to marry. Bride price raises the barrier to marriage in some places and among some groups. But it is one part of a much larger story.

Still, Sun makes one observation that deserves more attention than it has received. People may not withdraw because the price is high, he writes. They may withdraw because they cannot see where the price ends. That is the most important sentence in his letter. The problem is uncertainty.

Must there be an apartment before marriage? How much bride price is expected? Who pays for the wedding? Whose surname will the child take? Who stops working when the baby arrives? Who cares for aging parents? Does one spouse turn over his or her salary? How much are the parents on either side expected to contribute? If the marriage ends, what happens to the house, the savings and the money transferred before the wedding?

Traditional society once supplied answers—often unfair answers—to many of these questions. Those answers have weakened, but a new consensus has not fully replaced them. Young Chinese people have acquired much greater freedom in choosing how to live. They have also acquired much greater responsibility for negotiating almost every term of intimate life themselves. Freedom has expanded. So has uncertainty.

And this creates a peculiar equilibrium. Men fear becoming the family’s payer of last resort. Women fear becoming its caregiver of last resort. Men worry about open-ended financial obligations. Women worry about open-ended reproductive and domestic obligations. Both sides fear investing first. Both fear that the other person retains an easier exit. Both fear becoming the one left holding the risk.

Once those expectations take hold, delay becomes rational. Wait another year. Save a little more. Watch what happens to other marriages. Do not have the child yet. Do not register the marriage yet. Perhaps do not marry at all. That, more than any hypothetical national bride-price schedule, is the real “two hundred million people” problem.

What young people are waiting for is not a government-issued price tag for marriage. They are waiting for clearer boundaries.

Those boundaries are not difficult to describe. Bride price is not an entrance fee to marriage, and an extravagant bride price is not a man’s legal duty. Large transfers before marriage should be documented clearly. Each spouse retains full bodily and personal autonomy. Money, housework and childcare are shared responsibilities, though the precise division will differ from family to family. When a relationship ends, property disputes belong in court; intimate secrets should remain private.

And society has responsibilities too. If a country wants more people to marry and raise children, it cannot place nearly all the financial and professional risks of family formation onto two young adults and then wonder why they hesitate. Housing, childcare, education, parental leave and career interruption are not merely private problems. They shape national demographic behavior.

The goal should not be to make marriage cheap. It should be to make its risks more legible and its responsibilities more bounded. Then marriage might once again look less like an exercise in counterparty risk and more like what it is supposed to be: a way of building a life with another person.

So, Mr. Sun, if thirty years from now a young man comes to you and asks, “How much does a man have to pay to get married?” I hope you do not tell him 30,000 yuan, 300,000, three million or 30 million. Tell him something else: you are not buying a wife. She is not selling you a family through childbirth. Marriage is a decision by two free adults to make themselves partly responsible for each other’s future.

Money should be traceable. Rights should be explicit. Responsibilities should be reciprocal. And some things should remain beyond price.

The question young people most need answered is not: How much does marriage cost? It is: If I enter this relationship, are there limits to what I can be asked to give? Are my rights protected? Will the other person bear responsibilities too? And if one day we can no longer live together, can we still separate as two adults rather than destroy one another?

If a society can provide credible answers to those questions, whether the bride price is 30,000 or 300,000 yuan begins to matter much less. And if China is genuinely worried that young people are retreating from marriage and parenthood, the worst possible response would be to teach men and women to calculate against each other more aggressively. The better response is to build a society in which neither side has to fear that love is an unlimited liability.

Appendix: Justin Sun’s Open Letter to Hu Xijin

Dear Mr. Hu,

You ended your second article by saying that people should try to be honest and decent. So I would like to ask you something honestly.

I was born in 1990. I am thirty-six years old, at the age when marriage and children become real questions. You are exactly thirty years older than I am. I was already reading your articles when I was studying at Peking University. That is why I want to ask you rather than someone else.

There are about two hundred million men in China between the ages of twenty and forty, and about two hundred million women. I am not in a position to speak about the costs women bear. I will speak only about our side. We do not know how to build relationships with women, how to form families or how to raise children. We have never walked this road before. No one taught us how. For outsiders, these questions are gossip. For two hundred million young men, they are part of everyday life.

I do not know whether a Chinese man is expected to pay a large bride price. If he is, should it be a fixed amount or a percentage of his wealth? I do not know what price a man is expected to pay for having children. I do not know whether a man is expected to satisfy every request made by a woman before and after marriage. If he does not, and she leaves, does he have the right to recover the bride price? If she refuses to return it, may he go to court?

Nor do I know whether, after a man has observed all the expected formalities, met the parents, been separated for a long time, failed to recover his money and can no longer find the other person, he has the right to tell the public what he thinks. I chose to speak. Was that wrong?

I do not have answers to these questions. The state has answered some of them. The No. 1 Central Document addresses the issue, and the Supreme People’s Court has issued judicial interpretations. I have read them, and I trust them. But government documents are written on paper. Bride prices are negotiated around dinner tables. Between the two lies social reality.

You are my elder, and I respect your answer as well, just as I will unconditionally respect the court’s final judgment in the bride-price case.

A bride price has to be paid one way or another. North or south, educated or uneducated, the difference is largely whether it is 30,000 yuan, 300,000, three million or 30 million. This is a cost two hundred million young Chinese men are expected to bear. We accept that.

When a person cannot calculate the limits of a bargain, the safest choice is not to enter it. In 1990, the year I was born, more than 23 million children were born in China. Last year, there were 7.92 million—the lowest number since the founding of the People’s Republic.

I do not believe this entire decline can be explained by one calculation. But among the two-thirds who are now missing, many did not withdraw because they did not want marriage or children. They withdrew because they did not know what this road would ask of them, how much they would have to give, or whether what they gave would ultimately count.

People do not necessarily walk away because the price is high. They walk away because they cannot see where the price ends. We do not know the rules governing that price.

Two hundred million people are waiting for an answer, because this is the first step toward forming families, and it will help determine who makes up this country over the next thirty years. I do not have the answer today. I hope that thirty years from now, when a young person asks me the same question, I will be able to give him one.

Justin Sun

Ma Siwei

In the age of AI, one comforting phrase is often heard: everyone is back on the same starting line. This sentence is half true, and half dangerous.

The true part is that AI has indeed lowered many thresholds. In the past, an ordinary person who wanted to write a decent business plan had to know the format, do research, organize the language, and revise it again and again. Now AI can produce a framework in a few minutes. In the past, someone who could not code had a hard time building even a small tool. Now he can ask AI to generate scripts, web pages, spreadsheets, and automated workflows. In the past, writing, design, translation, market analysis, PowerPoint presentations, and résumé revision all required long training. Now many surface-level tasks have been flattened by tools almost at once.

This is where the idea of “the same starting line” comes from. Tools have become cheaper. Entry has become easier. For the first time, ordinary people have access to forms of assistance that once belonged mainly to professionals. A small shop owner can use AI to write advertising copy. A freelancer can use AI to prepare a client proposal. A student can use AI to simulate an interview. An ordinary worker can use AI to study an equipment manual. A crack has indeed opened.

But the danger lies here as well. The starting line may look the same, but the resources behind it are not the same. Some people own models, computing power, equity, data, platforms, and distribution channels. Others merely have a chat window. Some use AI to expand their companies. Some use AI to keep their jobs. Some use AI to train teams. Some use AI to help their bosses make their own labor cheaper. The tools look similar on the surface, but the positions are not similar at all.

So ordinary people in the age of AI should not be over-comforted by the phrase “the same starting line.” The real question is not whether everyone has a tool. The real question is who can use the tool to tell a different story, find a problem others have not seen, and turn that story into a plan that can actually land.

Peter Thiel’s thinking is useful here. His most famous question is: What important truth do very few people agree with you on? This sounds like a question from an investor, but it is also a survival question for ordinary people in the AI age. AI is especially good at producing “what most people have already said.” Ask it where the next big opportunity is, and it will easily give you an average answer. Ask it to write an industry analysis, and it will mix public information, mainstream views, and common judgments into one smooth document. It will be fluent, complete, and plausible. But very often, it is only the statistical product of imitative desire.

This is AI’s narrative trap. Human beings are already prone to imitating one another. When others say short videos make money, everyone rushes into short videos. When others say livestream commerce is an opportunity, everyone rushes into livestreaming. When others say AI applications are the next wave, everyone rushes into AI applications. Now AI makes this imitation faster. A person who was merely following the crowd can now follow the crowd more elegantly with AI. A company that was merely copying its competitors can now use AI to produce, at greater speed, a strategy that sounds advanced.

So competition becomes more crowded. This is not because everyone has become more creative. It is because everyone can produce a passable narrative faster than before. In the past, storytelling still had some cost. You had to know how to write, how to package, and how to prepare a proposal. Now the cost of narrative is close to zero. In one night, a person can generate ten business plans, twenty brand stories, and thirty short-video scripts. But when everyone can tell a story, the truly valuable thing is no longer the ability to tell one. It is whether what you are telling is something others have not yet seen.

This is what Thiel calls a “secret.” A secret is not a mysterious conspiracy. Nor is it some piece of inside information hidden from the public. It is more like a real problem that has not yet been accepted by the mainstream narrative. The secret Airbnb saw in its early days was this: strangers might be willing to stay in strangers’ homes. The secret Uber saw early on was this: in many cities, the problem was not simply a lack of cars, but a foolish way of matching vehicles, drivers, and passengers. These secrets later became business models. But before they were proven, they sounded like strange, counterintuitive remarks.

In the AI age, ordinary people also need to look for their own “secrets.” This secret does not have to be the founding of a unicorn company. It can be small. A teacher discovers that students are not unable to write essays; they have simply never gone through the experience of asking real questions. A nurse discovers that patients fear not only the test result, but the fact that no one explains it in language they can understand. A factory technician discovers that machine vision can detect defects, but does not know which kind of defect will become a major problem in a later process. A community worker discovers that elderly people are not unable to use smartphones; they are afraid that if they press the wrong button, no one will help them. A small shop owner discovers that customers do not only want low prices; they also want to be remembered.

These secrets come from the scene itself. AI can summarize existing knowledge on the Internet, but it is hard for it to walk into the scene for you. It can generate a nursing workflow, but it cannot see a patient’s fingers clutching the bedsheet. It can write an education plan, but it cannot hear the sense of inferiority in a student’s silence. It can analyze a market, but it cannot tell you why, at four in the afternoon, no one suddenly passes the door of a small shop. The advantage of ordinary people is often hidden in these places. It is not grand, but it is real. It may not be beautiful, but it is useful.

So the first survival rule for ordinary people in the age of AI is anti-imitation. Do not immediately ask AI to help you make a similar version the moment you see what someone else is doing. The more easily something can be copied by everyone, the faster it will lose value. First ask: Is this a story that has already been told to death? Is this demand real? Has this problem appeared in the actual scene? Is there a detail that the mainstream narrative has always ignored?

The second survival rule is to turn the “secret” into a small, verifiable plan. Simply saying “AI will change education” is not a story. It is empty talk. Saying “AI will help us improve efficiency” is not a story either. It is waste language. A real story must be specific. For example, a community school can use AI to do three things: help teachers generate tiered exercises, help students understand mistakes, and help parents read learning reports. Each task must be testable. Do students understand better? Do teachers spend less time on repetitive labor? Are parents less anxious? If it cannot be verified, the story is only a slogan.

The third survival rule is definite optimism. Definite optimism does not mean shouting every day that “the future will be better.” That is too easy. Real definite optimism means believing that the future can become better, and then offering a concrete path. In the AI age, the most useless sentence is “AI is important.” That sentence contains no information anymore. A more useful sentence would be: in the next three months, AI can help a certain group of people reduce a certain kind of low-value labor; it can reduce a certain kind of error in a certain workflow; it can give one ordinary person a certain new ability. The future must land in specific things. Otherwise, it is only motivational noise.

The fourth survival rule is to protect one’s own main brain. AI can help draft, compare, search for materials, and generate counterexamples. But the final judgment cannot be handed over. A person who knows only how to ask AI questions, but not how to judge AI’s answers, has not become stronger. He has only become easier to lead by polished language. The smoother the answer AI generates, the more slowly a person should move. Where might it be wrong? Is there a source? Is there a counterexample? Who benefits? Who bears the cost? These questions are not complicated, but they can save a life.

The fifth survival rule is to accumulate materials that AI cannot easily touch. The most valuable person in the future may not be the one who writes the best prompts, but the one who has real experience, real observation, and real problems. A person who has spent ten years in nursing, a person who has delivered food for five years, a person who has long adjusted machines in a factory, a person who knows the needs of elderly people in a community, a teacher who truly understands the psychology of students—all of them have their own materials. The problem is that in the past, they often did not treat these materials as knowledge. In the AI age, they must learn to speak these experiences out, organize them, and turn them into plans, courses, products, services, and public discussion.

The sixth survival rule is to treat AI as an exoskeleton, not as a substitute. An exoskeleton helps a person go farther. A substitute slowly hollows a person out. Using AI to write a first draft does not mean letting AI think for you. Using AI to do analysis does not mean letting AI make judgments for you. Using AI to make a plan does not mean letting AI face the consequences for you. Ordinary people need to learn a new rhythm of work: first raise the question themselves, then let AI expand it; first decide the direction themselves, then let AI gather materials; first write a rough idea themselves, then let AI help refine it; and at the end, revise, verify, and take responsibility themselves.

The real gap in the AI age will not appear only in whether people know how to use the tool. The deeper gap will appear here: some people use AI to imitate others, while some use AI to enlarge themselves; some use AI to produce clichés, while some use AI to test secrets; some use AI to chase trends, while some use AI to build their own direction; some use AI to submit a prettier assignment, while some use AI to discover a question no one has seriously asked before.

This is the true survival method for ordinary people. Do not imagine that the starting line is really fair, but do not give up simply because resources are unfair. Ordinary people do not need to compete with giants in computing power. They do not need to compete with models in speed. What they can truly do is protect their own scene, train their own judgment, find a small but real problem, connect AI to it, and slowly deepen it.

AI will make mediocre narratives cheaper and cheaper. It will also make real experience more precious. It will make following the crowd easier, and make resisting the crowd more valuable. It will allow most people to say the same thing faster, and it will reward the few who can say something that truly belongs to them.

In this sense, the most important competition in the AI age is not a race between humans and machines. It is whether a person can avoid being carried away by machines and the crowd at the same time. Are you imitating someone else’s story, or are you telling your own? This is the first question ordinary people must answer as they stand before the AI tsunami.

Subscribe now

BY：Du Hongwei （杜鸿威）

There is a popular “floor theory” of cognition. Some people, the idea goes, stand on the second floor and can see only the street in front of them; others stand on the eighth floor, where the view is broader and stretches much farther. Even when they are looking at the same world, they therefore reach very different judgments. The metaphor is vivid, and it does capture something real about life: differences in knowledge, experience, professional training, and sources of information can lead people to form very different pictures of the world. But the second-floor/eighth-floor metaphor also contains a danger. The moment someone says, “I’m on the eighth floor and you’re on the second,” a metaphor meant to explain cognitive differences quietly turns into a cognitive hierarchy. A harder question follows: Who decides which floor anyone is on? Could someone who is actually on the second floor be absolutely convinced that he is already on the eighth? If people assign their own floors, there is no reliable distinction between “I can see farther than you” and “I believe I can see farther than you.”

This is where the problem of cognition becomes genuinely interesting. People obviously differ in knowledge, experience, and ability. But what turns those differences into a gulf that is difficult to cross is often not how much a person knows today, but whether his cognitive system is still capable of revising itself. What matters most is not whether someone stands on the second floor or the eighth, but whether there is still a staircase between them.

Kant Never Issued Anyone an “Eighth-Floor License”

When people discuss cognitive differences, they often invoke Kant’s idea that “the human mind legislates for nature.” It is easy to take this to mean that people possess different cognitive capacities and therefore see different worlds—some more superficially, others more deeply. But that is not what Kant was really saying. Critique of Pure Reason asks a more fundamental question: How is human experience possible? How do space, time, and the basic categories of the understanding take part in our knowledge of the empirical world? Kant argues that what we can know are phenomena as they appear through the forms of human cognition; we cannot simply identify what we know with the “thing in itself.” More importantly, these forms of cognition are, in his account, common conditions that make human experience possible. They are not private cognitive operating systems, one installed in Zhang San and another in Li Si.

What Kant can genuinely contribute to the second-floor/eighth-floor debate, then, is not cognitive superiority but epistemic humility. Seeing the world is not the same as possessing the world as it is in itself. Knowing a great deal does not give anyone a God’s-eye view. Every judgment we make has already passed through perception, language, concepts, experience, and prior knowledge. The more clearly a person understands this, the less he should ask, “Why can’t other people see the view from my eighth floor?” and the more he should ask: How do I know that I am really on the eighth floor? One of the easiest mistakes for intellectuals to make is not ignorance, but turning their knowledge into a ruler with which to measure everyone else’s intelligence. The more one knows, and the more tools one has for explaining the world, the easier it can become to mistake one’s map for the world itself.

The old saying “You cannot speak of ice to a summer insect” is a beautiful metaphor. An insect that lives only through spring, summer, and autumn has never seen snow or ice, so it is hardly surprising if it does not believe in winter. The metaphor could lead us toward tolerance: human experience is always limited, and we should not immediately conclude that someone is stupid simply because he does not know what we know. Yet the metaphor can also slip almost imperceptibly in another direction. It begins with “everyone has limitations,” moves to “his limitations are greater than mine,” and ends with “I understand why he is wrong because I stand higher than he does.” What began as epistemic humility has become an epistemic hierarchy.

The deepest wisdom in “You cannot speak of ice to a summer insect” may not be that some people will never understand winter. What matters more today is another possibility: any of us may be that summer insect. Someone who sees political questions very clearly may become a summer insect when the subject turns to economics. An economics professor may be one in medicine. Even someone who has read Kant closely may become a summer insect when new facts appear, simply because he has become too attached to his theory.

True epistemic humility, then, does not mean standing on the eighth floor and feeling sorry for those on the second. That is still the arrogance of the eighth floor. Real humility means admitting that the view before us also has limits, that there may be a ninth or tenth floor we do not yet know exists, and even that what we once took to be the “eighth floor” may turn out to have been a mistaken mark on the map.

In the end, cognitive differences between people can be measured by two very different questions. One is: What floor are you on now? The other is: If new facts prove you wrong, are you willing to take the stairs? The first question easily produces a sense of superiority. The second determines how far a person, a science, an organization, or even a civilization can ultimately go. Genuine cognitive strength has never meant being right forever. It means preserving a capacity that is not very flattering, and can sometimes be painful: when you discover that you were wrong, you actually change.

The Greatest Danger Is Cognitive Self-Immunity

If someone says, “You disagree with me because your cognitive level is lower than mine,” meaningful discussion is already difficult. If he goes a step further and says, “The more you argue back, the more clearly it proves that your old programming has activated its defense mechanism,” the theory acquires a very dangerous property: it becomes almost impossible for it to fail. If you agree, that proves it is right. If you disagree, that also proves it is right. If you become angry, your programming has been triggered. If you calmly present facts, that can be interpreted as evidence that your programming is simply more sophisticated. Whatever happens can eventually be reinterpreted as proof that the theory itself was correct. A theory capable of explaining every possible outcome may look extraordinarily powerful, but in fact it violates Gödel’s incompleteness theorem and, more important, loses the one thing it most needs: it no longer knows what could prove it wrong.

Popper’s account of falsifiability has, of course, been revised and challenged in many ways, and science is far more complicated than the phrase “capable of being falsified.” Yet one simple epistemic principle remains useful: a claim about reality should at least allow for the possibility that some facts might conflict with it. Otherwise, what we have is no longer a judgment open to testing by reality, but a belief system designed to protect itself from reality.

This condition can be called cognitive self-immunity. The biological immune system identifies and rejects foreign substances; a cognitive system can develop a similar mechanism. Any information inconsistent with an established worldview is automatically recoded as “enemy propaganda,” “manipulation by vested interests,” “the other side has been brainwashed,” “their cognitive level is too low,” or simply “you don’t understand.” The result feels comfortable: one’s worldview can never lose. But the cost is enormous. A cognitive system that can never fail also loses the capacity to learn.

So there is a better question for judging someone’s cognitive level than “How much do you know?” It is this: What evidence, if it appeared, would make you change your mind? If the answer is “none,” it no longer matters very much whether the person is on the second floor, the eighth, or the eighteenth. The staircase has already been torn down.

What we really need to distinguish, therefore, is not “high-cognition people” from “low-cognition people,” but two different cognitive structures. One may be rich in knowledge yet closed. Such a person may have read widely, mastered a large vocabulary of concepts, possess strong logical skills, and be especially good at debate. Yet all that knowledge ultimately serves one purpose: proving that he was right all along. Another person may know less, but remain open. He habitually asks: Which parts are facts, and which are merely inferences? How strong a conclusion does this evidence actually support? Are there alternative explanations? If my prediction fails, how much should I revise my original model?

The first person may already stand high above the street, but have no staircase. The second may still be on the second floor, yet remain capable of climbing. Knowledge is therefore something like cognitive stock, while the capacity for correction is cognitive mobility. How much a person carries in his head certainly matters, but another question matters even more: Does he have a mechanism for discarding false knowledge, revising old models, and absorbing new evidence? This is why being “smart” is not the same as having strong cognition. Some intelligent people are exceptionally skilled at finding reasons for their own mistakes. The better their logic and the greater their knowledge, the more elaborate the fortress they may be able to build around their original conclusions.

The hardest cognitive skill is not proving that you are right. It is recognizing: This time, I may be wrong. A person can test himself with a few simple questions. Can I accurately state the strongest argument on the other side, instead of attacking only its weakest or most foolish version? Have I ever publicly revised a judgment because the facts changed? When a prediction fails, do I revise my model, or do I keep adding qualifications until I can once again conclude that “I was never really wrong”? These questions tell us more about the quality of a person’s cognition than how many books he has read or how many philosophers he can discuss.

The Core of Civilization: Can Errors Be Discovered, Transmitted, and Corrected?

Once the issue moves from individuals to institutions, something even more interesting appears. Three major institutions of modern civilization seem very different from one another: science, competitive markets, and democratic politics. They deal respectively with knowledge, resources, and power. Yet at a deeper level they share one premise: anyone can be wrong.

Scientists are not saints. They make errors of judgment, become attached to their own theories, and can be influenced by prestige, funding, academic schools, and personal interests. Science remains one of humanity’s most reliable systems for producing public knowledge not because scientists never make mistakes, but because science has developed mechanisms such as open methods, testing against evidence, peer criticism, replication, correction, and retraction. What makes science reliable is not that every paper is reliable, but that errors can, in principle, continue to be challenged.

Markets, under certain conditions, contain a similar logic. When Friedrich Hayek wrote about “the use of knowledge in society,” his most important point was not that entrepreneurs are smarter than government officials. It was that the knowledge required for a functioning society is dispersed among countless individuals and cannot be fully assembled in any single central mind. Where competition is able to operate, prices, profits, losses, and exit continually provide feedback. Entrepreneurs can certainly be wrong, but errors carry costs, while new participants remain free to propose another solution.

This does not mean that markets are always right. In Competition and Entrepreneurship, Israel Kirzner also makes clear that the permanent condition of real markets is not orderly equilibrium but ignorance, disequilibrium, and disorder. Sellers do not fully know buyers’ maximum demands; buyers do not know where lower prices may be found. Resources are misallocated, supply and demand become disconnected, and information barriers exist everywhere. These cognitive blind spots, shared by everyone in the market, are precisely where entrepreneurial profit arises. Every market participant sees only part of the picture and operates with incomplete information. Pure entrepreneurial profit, in this account, does not come from labor, capital, or technology as such, but from differences in perception.

This challenges the ordinary intuition that rewards simply follow labor or capital. Why do some people work from dawn to dusk all year and still remain close to subsistence? One answer is that they compete on tracks everyone can already see, exchanging time and physical effort for standardized compensation. Why do others appear to enter a field almost effortlessly and earn extraordinary returns? Because alertness allows them to notice cognitive blind spots others have missed and move into spaces where competition has not yet arrived.

For Kirzner, the essence of entrepreneurship is neither innovation nor risk-taking, but alertness. Schumpeter’s theory of innovation has been enormously influential, and generations have come to imagine the entrepreneur as a revolutionary who destroys an old order and creates something new, disrupting equilibrium through creative destruction. Kirzner supplied a different part of the picture. The two describe different commercial logics: Schumpeter’s entrepreneur disrupts equilibrium; Kirzner’s entrepreneur discovers and helps correct disequilibrium. Most ordinary people lack the power to reinvent an industry or the capital to reconstruct an entire market, but entrepreneurial alertness is not necessarily reserved for a tiny elite. Profit opportunities often do not have to be created from nothing. They may already exist, unnoticed.

None of this means that markets are omnipotent. Monopoly, externalities, asymmetric information, public goods, and badly designed incentives can all distort market feedback. The important question is therefore not whether “the market is always right,” but whether a system can expose errors and reallocate resources after failure.

Democratic politics follows a similar logic. The best defense of democracy has never been that “the majority is always right.” Majorities can make disastrous decisions. One of democracy’s more important virtues is that it can turn political mistakes into mistakes that are, in principle, correctable: policies can be criticized, information can enter the public sphere, those in power can face opposition, and governments that fail can be peacefully replaced through institutional procedures.

Seen from this perspective, science, competitive markets, and democracy belong to very different domains, but they share an important civilizational logic: they do not place their hopes in finding someone who will always be right. They try instead to prevent errors from becoming permanent. That may matter far more than deciding who is the smartest.

Return now to the second floor and the eighth. The problem looks different. Societies obviously contain real differences in expertise. Doctors generally know more about medicine than laypeople. Structural engineers know more about bridges than political commentators. Economists who have spent years studying monetary policy are usually better equipped than outsiders to understand inflation. Pretending that every opinion deserves equal professional weight is not epistemic humility; it is the abolition of expertise.

But expertise cannot automatically become an “eighth-floor license.” A brilliant economist may know almost nothing about viruses. A Nobel laureate may make terrible judgments once he leaves his own field. The more complex modern knowledge becomes, the less plausible it is that anyone can remain on the eighth floor of every subject.

So the real question is not whether different floors exist. It is whether there is still a staircase. Can new information enter? Can criticism be voiced? Can counterexamples be preserved? Can mistakes be made public? Can a theory be revised when its predictions fail? Can a policy be changed when it fails? Can those in power be replaced when they fail? The true cognitive capacity of a civilization is ultimately determined not by the brilliance of its smartest few, but by whether errors can be discovered, transmitted, and corrected within the system.

This leads to a very different conclusion from the conventional second-floor/eighth-floor metaphor: the greatest danger to a society is not that too many people are on the second floor. It is that the people on the eighth floor close the staircase and announce to everyone below: There is no need to come up. I have already seen the whole view.

Many of history’s greatest disasters did not occur because ordinary people knew nothing. They occurred because people with immense power and resources became convinced that they already knew enough—and then removed other people’s ability to correct them.

That is the most dangerous form of cognitive self-immunity: the moment it moves from an individual mind into an institution. When one person refuses correction, he may ruin an investment, a relationship, or a judgment. When a powerful institution refuses correction, its errors can be reproduced across an entire society.

In January 2025, a notice appeared at the matchmaking corner in Chengdu’s People’s Park: “Registration for women is full and temporarily suspended.” A photograph of the sign quickly spread online and became the basis for a striking story. China has more men than women, so why does a matchmaking corner seem to be full of women looking for husbands? Soon an even more compelling explanation emerged: men have stopped coming. They have done the math. A home, bride price, wedding, and children all cost money, while marriage also carries the risk of divorce. So some men simply opt out and spend their money on travel, gaming, fishing, skincare, and themselves. Men once earned money in order to support a family; now, the argument goes, they are learning to spend it on their own lives.

Subscribe now

The notice at People’s Park was real. On January 14, 2025, the park did temporarily suspend new registrations from women because the available spaces for their profiles had filled up. But park staff later told the media that the profiles actually displayed on the wall were still divided roughly fifty-fifty between men and women, and that there had also been occasions when the spaces for men had filled up first. A real notice was gradually enlarged into a national conclusion that the evidence did not support: the old saying that “a grown man should marry” had supposedly become “grown men no longer want to marry,” while the ranks of “leftover men” had somehow vanished, leaving only women unable to find husbands.

A 2023 special survey by the Center for Population and Development Studies at Renmin University of China, which examined people delaying or rejecting marriage and childbearing, found that only about one-fifth of unmarried or late-marrying respondents said they did not intend to marry in the future. In other words, most people who had not married had not rejected marriage altogether. Some 38.16 percent said they simply did not have enough time to date or get married. That finding matters because it brings the issue back from the realm of values and attitudes to the realities of daily life.

A young person may remain unmarried because he or she does not believe in marriage. But the reason may also be long working hours, unstable income, the lack of affordable housing, failure to meet a suitable partner, or the feeling that starting a family is simply beyond one’s current means. “Not marrying” includes both active refusal and involuntary delay. Chinese public discussion often blurs the two. If a man remains single, people say he has “given up.” If a woman remains single, they say her standards are too high. It is a convenient explanation because it puts responsibility back on the individual and makes the social structure disappear.

What is actually happening to marriage in China is far more complicated—and more consequential—than the claim that “men no longer want to marry.”

Why Do Women Still Feel That Men Are Hard to Find?

Start with the figure most easily misunderstood. China’s 2020 population census recorded 723.34 million men and 688.44 million women, a difference of roughly 34.9 million. The number is real. But those 34.9 million are not 34.9 million single men waiting for wives. They include children, elderly people, married men, widowers, and people living in different regions and belonging to different age groups. An excess of seventy-year-old men means almost nothing to a thirty-year-old woman looking for a partner. What matters in a marriage market is not the total number of men and women, but the number who can realistically match in age, place, education, income, and marital status.

The overall gender gap is also narrowing. According to the National Bureau of Statistics’ 2025 one-percent population sample survey, China had about 717.22 million men and 688.23 million women, reducing the gap to roughly 29 million and bringing the overall sex ratio down to 104.21 men for every 100 women. The more serious imbalance lies among people of marriageable age. One study based largely on 2020 census data estimated the sex ratio in China’s marriageable population at roughly 110 men for every 100 women. It also found 17.33 million never-married men aged thirty-five or older, around 8.82 million of them in rural areas. Rural men faced a substantially higher risk of remaining unmarried for life than urban men.

China, then, does not suffer from a nationwide “surplus of women.” In aggregate demographic terms, the opposite remains true. Men still outnumber women, and the burden of that imbalance falls disproportionately on men with lower incomes, less education, and fewer opportunities in regions losing population. This is the first marriage problem China faces: many men have difficulty finding wives.

And yet many urban women also genuinely feel that there are few suitable men around them. How can both things be true? Because a marriage market is not a stadium into which ten million men and ten million women can simply be poured and paired off one by one. Marriage is a matching process. Equal numbers do not guarantee compatible matches.

Over the past four decades, China has undergone three enormous social changes: mass migration from the countryside to the cities, a dramatic rise in women’s educational attainment, and major changes in household income and occupational structure. Marriage expectations, however, have not changed at exactly the same speed.

Traditional Chinese mate selection has long shown a broad tendency for men to marry women who are similar to them, somewhat younger, or somewhat lower in educational and economic status, while women have tended to prefer men whose age, education, occupation, and economic position are at least equal to their own. In the past, this system could operate relatively smoothly because men on average had more education and greater economic resources than women. The educational revolution has changed that structure. As more women attend university and enter professional careers, the old marriage hierarchy begins to malfunction. If women’s education and occupational status rise while the expectation that “the man should not rank below the woman” remains largely intact, a peculiar pattern emerges: more men accumulate at the lower end of the marriage market, while more women accumulate at the upper end. Both groups feel that suitable partners are disappearing.

A 2025 economic study of China’s marriage market found that educational expansion over the past two decades has itself contributed substantially to falling match rates. Changes in educational structure have affected both men and women, in part because highly educated women and less-educated men increasingly occupy opposite sides of a structural mismatch. The study found little evidence that highly educated women had begun marrying substantially less-educated men in large numbers. Earlier national research had already identified a similar pattern: men with low levels of education and women with high levels of education were among the groups less likely to enter marriage.

This produces one of the most revealing—and most misunderstood—scenes in contemporary China. Rural men say there are no women. Urban women say there are no men. Both may be describing their own reality accurately. The rural man means: there are too few women who are willing to stay here and marry someone like me. The urban woman means: there are too few men whose education, career, lifestyle, and views of family fit what I am looking for.

Tens of millions of statistically “extra” men do not automatically solve the marriage problem of a thirty-five-year-old female doctor, lawyer, or engineer in Beijing, Shanghai, or Chengdu. Nor will a low-income forty-five-year-old bachelor in a western county automatically marry a thirty-five-year-old highly educated single woman in Shanghai simply because both happen to be unmarried. They live in the same country, but not necessarily in the same marriage market. The so-called “leftover men” and “leftover women” are often the two ends left behind by the same structural mismatch.

Marriage Is Becoming Less a Life Stage Than a High-Cost Decision

Economics offers one of the most important keys to understanding marriage in China today. In the 2023 Renmin University survey, some respondents looking back on their wedding expenses reported an average cost of 330,400 yuan—more than eight times China’s per capita disposable income of 39,200 yuan that year. The figure must be used carefully. It came from a specific survey and should not simply be presented as “the average cost of getting married in China.” Nor does it mean that men alone pay all 330,400 yuan. In reality, marriage expenses may be divided among the groom’s family, the bride’s family, and the couple themselves, with enormous differences between regions.

Still, the number captures a real shift: the price of entering marriage has risen sharply. Housing, wedding ceremonies, bride price, renovation, and cars may only be the beginning. After marriage come mortgages, childbirth, childcare, education, and the care of aging parents. According to a 2024 estimate by the YuWa Population Research Institute based on official statistics, the average direct cost of raising a child from birth to age seventeen was around 538,000 yuan, rising to roughly 680,000 yuan through completion of an undergraduate degree. These are not official national standards, and individual families may spend much less or much more. But they illustrate the scale of the long-term financial burden facing young households.

Marriage has therefore undergone a profound change. For many people in the past, the main question was simple: I have reached the right age, so I should marry. Increasingly, young people now ask a different set of questions. Can I afford to marry? Will my life become better after marriage? Do I want children? Who will care for them? Who will pay the mortgage? Who will do the housework? Who will take care of both sets of aging parents? If there are no satisfactory answers, postponing marriage—or even declining it altogether—becomes understandable.

This does not mean that men suddenly became selfish or that women suddenly became demanding. Marriage is gradually changing from an almost compulsory stage of life into a long-term contract whose costs and benefits people feel entitled to examine.

The financial pressures on men are easy to see. The costs borne by women are often less visible. For many professional women, the greatest cost of marriage may not be the wedding itself, but what happens to their time afterward. Who becomes pregnant? Who gives birth? Who takes time off when a child is sick? Who attends school meetings? When an elderly parent needs care, whose working day gets cut short?

If a society expects women to have careers while continuing to assume that women will perform most family care, then the more education and occupational opportunities women gain, the greater the opportunity cost of marriage may become. This helps explain why rising female economic independence does not automatically produce higher marriage rates. On the contrary, economic independence gives women more power to reject poor marriages. In the past, an unmarried woman might face severe economic and social risks. Today she can work, rent or buy a home, travel, and maintain a social world of her own. Marriage changes from something she “must have” to something worth entering only if it improves her life.

Men are undergoing a related shift. For generations, the expected male life course was clear: earn money, buy a home, marry, have children, support the household. Some men now ask a different question: if my income barely supports my own life, why must I take on decades of housing costs and family obligations? It is misleading to call this a “retreat from responsibility.” A more accurate description is that family responsibility is shifting from a social command toward an individual choice.

The change affects both sexes, though not in exactly the same way. Men tend to be more sensitive to the financial threshold for entering marriage; women tend to be more sensitive to marriage quality, division of domestic labor, and the career costs of childbirth. Both sides are raising the conditions under which they are willing to marry. The market consequently becomes narrower.

Marriage and Childbearing: A Problem Chinese Society Cannot Avoid

China recorded only 7.92 million births in 2025. Marriage and childbearing remain closely linked in Chinese society, so a sustained decline in marriage will eventually affect births, household size, housing demand, schools, the pension system, and local government finances. Marriage may begin as a private matter between two people, but its consequences eventually spread through the entire society.

If policymakers understand the problem only as “young people do not want to marry,” they are likely to prescribe the wrong medicine. More matchmaking events will not solve unstable employment among rural men. Telling women to lower their standards will not address professional women’s concerns about the division of household labor. Criticizing men for “lying flat” will not make housing and childcare cheaper. Restricting excessive bride prices may reduce marriage costs in some regions, but it cannot eliminate a long-standing numerical imbalance between men and women. A few hundred yuan in marriage subsidies cannot offset the cost of a home, twenty years of child-rearing, and decades of family responsibility.

Effective marriage policy therefore eventually becomes social policy: reduce housing pressure on young adults, improve employment prospects, expand affordable childcare, reduce excessive educational competition, lower the cost of raising children, bring fathers more fully into childcare and eldercare, reduce the career penalty women face for having children, and build stronger pension and social-support systems for the large number of rural men who may remain unmarried throughout their lives.

At a deeper level, society has to accept a simple fact. The state can lower the barriers to marriage, but it cannot order one person to love another. Once a modern society gives people the ability to leave or reject low-quality marriages, it cannot fully return to an era when reaching a certain age automatically meant getting married. That is not necessarily a bad thing. The difficulty is that marriage itself has changed rapidly in China, while housing, employment, gender roles, eldercare, childcare, and expectations about suitable partners have not all changed at the same pace.

That is why China now presents such an apparently contradictory picture. There are still tens of millions more men than women, yet women in major cities often say that suitable men are difficult to find. Some matchmaking corners fill up with female applicants, while large parts of rural China still contain substantial numbers of middle-aged men who cannot find wives.

The title of this essay—Men Struggle to Marry, Women Struggle to Find Husbands—seems at first to describe two separate stories. Look more closely, and they are really one. China does not lack men, nor does it lack women. What China increasingly lacks are matches in which two people can accept one another and have the material conditions to build a life together.

In the past, marriage mainly answered the question of whether one had a spouse at all. Society defined the proper age, families supplied resources, and individuals largely followed an established life course. Today the question is increasingly whether marriage is worth entering. Men calculate income, housing, and family obligations. Women calculate careers, housework, and the costs of childbirth. Both have more freedom than previous generations to walk away, and both face a higher price for walking in.

What is changing in Chinese marriage, then, is not that men have suddenly “woken up,” or that women have suddenly become too choosy. An ancient institution is confronting, on a mass scale, a modern individual’s question: If being together does not make life better for both of us, why must we marry? And if a couple’s income cannot support the cost of raising a child, why must they have one?

That is the question behind the figures—6.10 million marriages, 6.76 million marriages, or 3.27 million marriages in half a year—that Chinese society can no longer avoid.

Subscribe now

Mr. Tang Shizeng has passed away.

On August 23, 2026, the journalist who had spent decades moving through wars, wildernesses, deserts, plateaus, and distant corners of the world died in Beijing after an illness. He was sixty-five.

I have read many of Tang Shizeng’s reports over the years. What impressed me most was not simply how many world leaders he had interviewed, how many dangerous places he had visited, or how many precious photographs he had taken. It was the almost stubborn journalistic spirit that defined him: “As long as I live, I will keep making news.” Few sentences could better sum up his life.

He was not a journalist who sat in an office waiting for news to arrive. He kept going out to meet the world where it was happening. From the Great Wall, Kekexili, and the Qinling Mountains to Iraq and the battlefields of the Middle East; from Shennongjia to Everest Base Camp; from Pakistan, Afghanistan, India, and Nepal to the old battlefields of the Second World War across Europe, Asia, and Africa, he continually placed himself in the middle of the real world.

For a war correspondent, news is never only words and photographs. It also means danger, exhaustion, illness, loneliness, and sometimes death.

Tang spent long periods reporting from the Middle East. Artillery fire, political upheaval, radiation exposure, harsh living conditions, and the enormous physical strain of working in war zones did not stop him from doing his job. In later years, he suffered from serious illness, and his health remained badly affected for a long time. Yet his urge to travel, his instinct to record, and his consciousness as a journalist never disappeared. That is something I especially admired in him.

It is not so unusual for someone to take risks when he is young, healthy, and full of energy. What is much rarer is to remain curious about the world after illness, aging, and physical suffering arrive—to remain willing to set out again, and still feel compelled to record the age in which one lives.

Tang Shizeng possessed an extraordinary capacity for hardship. He could enter a war zone, drive vast distances, travel alone, live under difficult conditions, and endure levels of exhaustion and risk that most people would prefer to avoid. For him, journalism was never a comfortable profession. It was a way of life.

His resilience was not confined to the battlefield.

From the 1990s onward, his health was seriously damaged. He endured years of illness and treatment, yet continued to interview people, take photographs, travel, and write. Even the titles of his books—I Returned from the Battlefield, Return to Baghdad, My Normandy, and A Journey Alone—seem to trace the rhythm of his life: setting out, returning, and setting out again. Even toward the end of his life, that impulse to record never left him.

Gao Yu once recalled an episode from 2021. When Zhao Ziyang’s family was about to move out of No. 6 Fuqiang Hutong in Beijing, where they had lived for thirty-two years, she asked Tang to go there and preserve a final visual record of the house. Tang rode over on his bicycle carrying nothing more than a digital camera. Using a continuous, unbroken shot, he recorded the courtyard, the study, the bedroom, the furniture, and the traces of the life Zhao Ziyang and his family had once lived there. That small episode says a great deal about the kind of journalist Tang Shizeng was.

News does not always happen under artillery fire. Sometimes an old courtyard about to disappear, a wooden bed, or an ordinary room is itself history. A journalist’s responsibility is to preserve it before it vanishes.

Tang Shizeng once said, “As long as I live, I will keep making news.” It was more than a bold declaration. It was a professional ethic.

For a true journalist, to be alive means to observe. It means to ask questions, to go to the scene, and to record what is real. The body may grow exhausted. Conditions may be harsh. Life itself may even be placed in danger. But the pursuit of facts cannot easily be abandoned.

Today, Tang Shizeng can no longer lift his camera or set out on another journey. But the reports, photographs, and words he left behind—and the journalistic spirit behind them, his willingness to endure hardship, confront danger, and remain faithful to the scene—are still worth remembering.

A journalist’s life is finite. A truthful record can outlive the person who made it.

As long as I live, I will keep making news.

May these words be the best way for us to remember Tang Shizeng.

With deepest sorrow, we mourn Mr. Tang Shizeng.

August 29, 2026
Nuremberg

Subscribe now

On August 26, 2026, Bill Gates (比尔·盖茨) disclosed a plan that has not yet been finalized: he hopes to visit China later this year and meet once again with Xi Jinping (习近平). The facts need to be clear from the start. A “Gates-Xi meeting” has not been confirmed. Reuters reported that Gates’s team is arranging his next trip to China and trying to secure a meeting with Xi, tentatively in November. As of the time of the report, China’s Ministry of Foreign Affairs had not confirmed such a meeting. Nor does Gates want to discuss Microsoft’s business in China. He wants to talk about artificial intelligence. He worries that increasingly powerful AI systems may be able to launch cyberattacks, design dangerous molecules, and assist in biological attacks, while also displacing large numbers of workers and fostering psychological dependence. In particular, he hopes the United States and China can establish some common restraints on the release of dangerous models and on AI capabilities related to biological security.

The two men have met before. On June 16, 2023, Xi received Gates in Beijing (北京) and called him “the first American friend” he had met in Beijing that year. Their discussion then focused on public health, poverty reduction, agriculture, and technological cooperation. Three years later, if they sit down together again, there will be a very different technology on the table. Public health requires cooperation between states. AI is more complicated, because the force driving it has come first not from governments but from companies. That is what makes this possible visit so interesting: a man who emerged from one of the great success stories of American commercial capitalism is now trying to persuade one of the world’s most powerful states to exercise greater control over technology. And he believes that it is not enough for China to regulate itself. The United States must act as well.

That may sound contradictory, but it is not. AI is changing a relationship between the state and capital that has been taking shape for centuries.

Why Gates Wants to Go to Beijing

What Gates is proposing now goes far beyond what we usually call “tech ethics.” In a long essay published on August 26, he argued that existing state institutions were not designed for the age of AI. Education departments oversee education. Labor departments deal with employment. Financial regulators supervise finance. Defense departments handle security. But AI is moving into all of these areas at once. Gates therefore calls for new systems of governance at both the national and international levels. Internationally, he suggests drawing lessons from nuclear inspection regimes, international aviation rules, and agreements that protect the ozone layer, though none can simply be copied. More important, countries that possess leading AI companies and control critical supply chains need to establish common rules early. Gates has said explicitly that this cannot be done without cooperation between the United States and China.

There is already something like private diplomacy in this. Gates is not an envoy of the American president, and he has no authority to negotiate on behalf of the U.S. government. But that is precisely what makes the question interesting. Why does a private entrepreneur and philanthropist believe that he should be discussing with China’s top leader a technology that may affect war, employment, social structure, and even human survival? The answer is not hard to find. Much of America’s most important AI capacity is already in private hands.

OpenAI, Anthropic, Google, Meta, and xAI are not government agencies. Neither is Nvidia. Nor are the vast cloud-computing infrastructures operated by Amazon, Microsoft, and Google. In 2025, more than 90 percent of the world’s notable frontier AI models came from industry rather than governments or universities. Private AI investment in the United States reached $285.9 billion that year, compared with about $12.4 billion in China—a difference of more than twenty-three to one. Measured by private capital alone, the American advantage is extraordinary. This gives American AI power a distinctive character: the government holds sovereign authority, but companies control more and more of the capabilities that matter.

This is also the question Sharon Weinberger (莎伦·温伯格) pursues in Valley of Death: How the Tech Titans Are Igniting the Future of War (《死亡谷：科技巨头如何点燃未来战争》).

Valley of Death looks at SpaceX, Palantir, Anduril, and a new generation of defense-technology companies. Weinberger is not mainly interested in who has built the cleverest drone or which Silicon Valley entrepreneur is the smartest. Her real subject is a shift taking place inside American state power. The old military-industrial system also depended on cooperation between government and business, but its basic logic was relatively clear: the state defined the need, the state supplied enormous amounts of research funding, major defense contractors built the weapons, and the government maintained substantial control through procurement, contracts, secrecy, and long-term programs.

A different kind of company has now emerged. Venture capital puts up the money first. Companies develop the technology themselves and retain the software, data, algorithms, designs, and much of the intellectual property. Once the product matures, the Pentagon becomes the customer. Weinberger describes this emerging relationship as a new “Pentagon–Silicon Valley complex.” Her warning is pointed: if satellite communications, AI analysis, cloud computing, unmanned systems, and perhaps future autonomous weapons become increasingly dependent on a handful of private companies, then who truly controls the technologies on which the state depends to fight a war?

This is not an abstract question. SpaceX has become deeply embedded in America’s national-security infrastructure. Palantir’s software is now used in intelligence and military systems. Anduril was built from the beginning around drones, sensors, and autonomous weapons. Silicon Valley once imagined itself mainly through the consumer Internet and the promise to “change the world.” Now it is moving back into the machinery of war. This is why one of the central insights of Valley of Death matters: we should not assume that start-ups are inherently more innovative than bureaucracies, nor should we imagine that “technology companies” and “the state” occupy two separate worlds. What matters is where the money comes from, who owns the technology, whom the state depends on, and, in the end, who has the power to say no.

That offers a useful way into the AI question. The United States may have the most powerful AI companies in the world, but that does not automatically mean that the U.S. government has the strongest ability to control AI. Those are two different things.

America vs. China: Capital Power vs. Organizational Power

What is hardest to reproduce about the American AI system may not be any single model. It is capital. A young engineer can leave Stanford, Berkeley, MIT, or Google, start an AI company with almost no revenue, and still raise hundreds of millions or even billions of dollars if investors believe that company might become the next major platform. Ten companies can fail, and the portfolio can still succeed if one produces the next OpenAI, Anthropic, or Nvidia.

The system can look astonishingly wasteful, but it is unusually good at exploring unknown technologies because real breakthroughs are difficult to plan in advance. Governments can write five-year plans, but they cannot reliably decide five years ahead of time which model architecture will win. Capital markets work differently: if I do not know who will win, I can place bets on many contenders at once. This is one of America’s deepest advantages in AI.

Stanford’s 2026 AI Index makes the point clearly. In 2025, 1,953 American AI companies received new funding, more than ten times the number in the country that ranked second. The United States also continued to produce more leading models: of the notable models released in 2025, 59 came from the United States and 35 from China. America also has vast data-center capacity and some of the world’s strongest cloud-computing companies. Put differently, the American system is very good at one thing: turning uncertainty into something capital can bet on.

Let a thousand ideas emerge. Let investors decide. Let talent move. Let companies fail. Let the winners take enormous rewards. It is a harsh system, and it creates bubbles and extreme concentrations of wealth. But it is well suited to the first half of a technological revolution, when the main challenge is not deployment but finding the right answer when no one yet knows what the right answer is. China’s institutional advantage lies somewhere else.

If we looked only at private AI investment, China and the United States would seem to be fighting in different weight classes. Yet their model capabilities are nowhere near twenty-three to one apart. In fact, Stanford’s 2026 AI Index reached a striking conclusion: the performance gap between the leading American and Chinese AI models has essentially closed. Since 2025, models from the two countries have repeatedly traded the lead. By March 2026, Anthropic’s strongest model held an aggregate performance advantage of only about 2.7 percentage points over the leading Chinese model. China also leads in the number of AI papers, citations, patents, and industrial robots installed.

That tells us something important. Money matters, of course, but money is not everything. Behind Chinese AI stands another kind of machine. National funds, local governments, universities, state-owned enterprises, Internet companies, manufacturing supply chains, electricity systems, telecommunications networks, data centers, and a vast pool of engineers can all be directed, to varying degrees, toward a common industrial objective.

In 2025, China’s State Council issued the Opinions on Deepening the Implementation of the “AI Plus” Initiative (《关于深入实施“人工智能+”行动的意见》). Its goal was not simply to build a few larger models. It was to push AI into science, manufacturing, agriculture, services, consumption, health care, and public governance. The document set a target for new-generation intelligent devices and AI agents to reach a penetration rate of more than 70 percent by 2027 and more than 90 percent by 2030. It also called for investment in infrastructure, data systems, and open-source ecosystems, backed by public funding, government procurement, long-term capital, and “patient capital.” By the end of 2025, China’s National AI Industry Investment Fund had reached 60 billion yuan. The 2026 government work report then called for the construction of ultra-large intelligent-computing clusters, better coordination between computing and electricity, nationwide monitoring and scheduling of computing resources, and the large-scale commercial use of AI in key industries.

None of these measures, taken by itself, guarantees technological leadership. Together, however, they create another kind of capability: the ability to move technology rapidly from the laboratory into the wider economy. America is especially good at moving from zero to one. China is becoming increasingly good at moving from one to one hundred. This is only a broad structural distinction, and it should not be turned into a rigid national stereotype. China also has highly original companies, while the United States has DARPA, NASA, the Department of Defense, national laboratories, and its own industrial policies. The real point is that the two systems place their weight in different places.

Why DeepSeek Makes America Uneasy

This is why the significance of DeepSeek (深度搜索) cannot be understood only in terms of model performance. If China had simply spent the same amount of money, bought the same chips, copied the same path as American companies, and produced a somewhat cheaper model, the United States would not have been nearly as alarmed. The real concern is different: China may have found another technology-cost curve even while facing constraints on capital and access to the most advanced chips.

Once the gap in model capability narrows, the center of competition shifts. The question is no longer who scores two points higher on a benchmark. It becomes who can put AI into factories, robots, cars, hospitals, government services, power grids, and supply chains at lower cost. At that point, decades of Chinese manufacturing experience suddenly become highly relevant.

If AI remains forever inside a chat box, the American advantage is enormous. But once AI grows hands and feet—once it moves into cars, drones, robotic arms, factories, electrical systems, and urban infrastructure—China’s integrated industrial base becomes another kind of computing power. This is something many discussions of the U.S.-China AI competition miss: computing power is not only GPUs. Once AI enters the physical world, it needs electricity, chips, data centers, sensors, motors, batteries, robots, manufacturing, maintenance, and supply chains. In the end, it needs an industrial society. So the second half of the AI revolution may come to look more like the Industrial Revolution than the Internet revolution. That is precisely the arena China wants to enter.

But one more step in this argument leads quickly to another trap: China has strong manufacturing and an effective state, therefore China will win. It is not that simple. State mobilization has costs of its own.

A government may know that it wants to develop “artificial intelligence,” but that does not mean it knows who should invent the next generation of AI or what form that invention should take. Subsidies can relieve shortages of capital, but they can also create large numbers of companies chasing subsidies. Local governments can build computing centers, but some of those centers may have little real demand. Public funds can provide patient capital, but if officials are afraid of being blamed for failure, they may steer money toward the safest firms, the best-connected firms, or simply the firms that most closely resemble the language of government policy.

Innovation has an awkward characteristic: the most important ideas often do not look like the “right answer” before they succeed. The state is especially good at concentrating resources on problems that have already been defined. Yet one of the hardest parts of innovation is defining the problem in the first place.

This is the “valley of death” that the Chinese model itself must cross. The state can provide money, but the market cannot disappear. Without real elimination, without companies dying, without capital taking losses, and without users being able to say no, what the state ends up with may not be innovation at all. It may simply be an impressive industrial plan. What China needs, then, is not for the state to replace capital. Quite the opposite. The state needs to solve problems that private capital will not solve, while capital must continue to perform one of its most important functions: discovering mistakes.

America Is Becoming More “Chinese”

The reverse is also true. The United States is hardly defending some pure version of the free market. In 2025, the White House released America’s AI Action Plan (《美国AI行动计划》), which treated artificial intelligence from the outset as a contest over global leadership. The plan set out three broad priorities: accelerating innovation, building AI infrastructure, and using diplomacy and security policy to strengthen America’s global position in AI. The government called for faster construction of data centers, semiconductor facilities, and energy infrastructure, while seeking to make American chips, models, and applications the global standard.

This is no longer a story of “government staying out of the market.” The American government is intervening, and it is intervening more deeply. Chip export controls are state action. Reforms to speed up approval of AI infrastructure are state action. Energy policy is state action. Defense procurement is state action. International technology alliances are state action. So the old shorthand—China relies on the state, America relies on the market—fits reality less and less well.

A more accurate description is this: the United States is adding state power to capital, while China is adding capital competition to a state-led system. The two paths are moving toward the middle. The real question is which country can find the better balance.

The deeper problem Weinberger raises in Valley of Death is not simply that American technology companies are entering the military sphere. It is this: as the state becomes increasingly dependent on companies, and as the technologies controlled by those companies begin to resemble forms of state power, who constrains whom?

SpaceX can provide satellite communications. Palantir can analyze battlefield data. AI companies may eventually determine which kinds of models can be used by militaries, intelligence agencies, or even nuclear command systems. This creates a relationship between state and capital that has little precedent: the state cannot do without capital, while capital increasingly controls capabilities that once belonged almost exclusively to the state.

That is why Gates’s desire to meet Xi Jinping is so symbolic. Gates is not a president, a secretary of state, or the secretary-general of the United Nations. He is a private citizen. Yet his wealth, technological networks, global influence, and access to expertise allow him to try to discuss with China’s top leader how humanity should govern a technology that may pose global risks. A century ago, that would have been difficult to imagine. Today it seems almost natural. That alone tells us something about how the relationship between state and capital is changing in the age of AI.

China’s AI Global Proposal

Gates has called for a new international organization to govern AI, but he is not speaking into an empty space. In July 2025, China released the Global AI Governance Action Plan (《人工智能全球治理行动计划》), emphasizing safety and controllability, respect for sovereignty, open cooperation, and inclusive development. China later proposed the creation of a World Artificial Intelligence Cooperation Organization and suggested that its headquarters could be located in Shanghai (上海).

If Gates does meet Xi in November, then, the two already have at least one subject on which a conversation can begin: AI cannot be left entirely without international governance. But agreement on the problem does not mean agreement on the answer.

Gates is most concerned about the loss of control over advanced models, biological risk, cyberattacks, and disruption to employment. His approach therefore leans toward safety thresholds and international verification for frontier capabilities. China’s public governance framework, by contrast, also stresses national sovereignty, the right to development, broad access to technology, and the need to narrow the digital divide. There is overlap here, but there are real differences as well.

The United States worries that advanced capabilities may spread too quickly. Many developing countries worry that those capabilities may remain forever in the hands of a small number of American companies. China stresses broad access; the United States stresses safety boundaries. But full openness may allow dangerous capabilities to spread, while total restriction could turn AI into a privilege reserved for a few countries and companies. That may prove to be one of the hardest questions in the international politics of AI.

Only now can we return to the question in the title. If an answer has to be given today, mine would be this: in 2026, the United States is still the world’s leading AI power overall, but its lead is no longer as secure as it looked three years ago. China is the only country approaching the United States at the same time in models, industry, talent, infrastructure, and the state’s capacity to organize resources.

America’s advantages are real. It has the world’s strongest private capital markets, the densest concentration of leading AI companies, first-rate universities, a global talent network, a lead in advanced chip design, and global cloud platforms such as Microsoft, Amazon, and Google. China’s advantages are equally real. It has a vast engineering workforce, an integrated manufacturing base, a huge domestic market, the ability to build infrastructure quickly, a government capable of coordinating resources, and an increasingly competitive open-source model ecosystem.

But the next stage of the competition may be decided less by these strengths than by each country’s ability to correct its own weaknesses. America’s central problem is whether a state can remain in control when capital becomes too powerful. China’s central problem is whether capital and companies will retain enough room to make mistakes when the state becomes too powerful.

The United States must answer this question: if a small number of technology companies control computing power, models, cloud platforms, satellites, and military software, how does a democratic state retain ultimate authority? China must answer another: if the state defines priorities, allocates capital, builds infrastructure, and opens markets for deployment, how can companies preserve a genuine willingness to take risks and think independently? The questions appear different, but they point to the same principle: innovation needs freedom; scale needs organization.

Freedom without organization can leave technology in the hands of a few powerful corporations. Organization without freedom can turn technology into another set of targets to be fulfilled. The strongest AI country will not simply be the one with the largest state or the most powerful capital. It will be the country best able to accommodate both forces at once: allowing capital to discover the unknown, companies to experiment and fail quickly, and universities to produce unconventional ideas, while giving the state the capacity to build infrastructure that private capital will not build, bear long-term risks, maintain safety boundaries, and apply the brakes when technology genuinely threatens society.

Gates’s Beijing Trip Is Really About the Next World Order

Seen from this angle, the still-unconfirmed November visit could mean far more than a headline announcing that “Gates will meet Xi Jinping.” The most dangerous technology of the twentieth century—nuclear weapons—was controlled mainly by states. One of the most powerful technologies of the twenty-first century—artificial intelligence—has emerged first in the hands of companies. That is a fundamental change.

The nuclear age asked: how can states constrain other states? The AI age adds another set of questions: how do states constrain companies, how do companies influence states, and how can different states jointly constrain a technology that does not respect national borders?

Gates’s life happens to span this change. As a young man, he believed software could change the world. In middle age, he demonstrated that a private company could accumulate economic power greater than that of many countries. Now, later in life, he has begun to worry that technology may be changing the world faster than political institutions can govern it. This is more than one man changing his mind. It may be a problem that Silicon Valley as a whole is about to confront.

Valley of Death puts a related question before us. Once technology companies move into war, national security, and critical infrastructure, we can no longer think of them as ordinary companies. They are beginning to acquire forms of power that once belonged to states. China is approaching the same intersection from the opposite direction: the state is moving into AI, while in the United States AI companies are moving deeper into the state. One path brings the state toward capital; the other brings capital toward the state. In the end, both arrive at the same question: who controls AI?

That is why the question of who leads in AI cannot be answered by the next model ranking. The true leader may not be the first country to build the smartest model, or even the country that spends the most money. It may be the first country to create a stable system in which innovation, capital, state power, security, and the interests of ordinary people can coexist over the long term.

A six-month lead in models can be erased. A one-generation gap in chips may also be closed. A shortage of capital can be overcome. Institutions are much harder to catch up with.

If we must draw a finish line for the U.S.-China AI competition, I would put it there: which country can make machines smarter without allowing society to lose control of them; allow capital to take risks without giving capital power that cannot be checked; and give the state enough capacity to organize the future without allowing the state to decide everyone’s future for them.

Whoever gets there first will be the real leader in AI.

Subscribe now

Li Zhide

On August 26, 2026, Bill Gates did two things worth considering together. First, he published an essay of nearly 6,000 words with an unusually weighty title: The Turbulent AI Era Is Here. The Choices We Make Now Matter. Second, Reuters reported that his team was arranging another visit to China later this year and hoping to secure a meeting with Xi Jinping, tentatively in November.

The factual boundary needs to be clear at the outset. There is no confirmed Gates-Xi meeting. Reuters reported that Gates’s team was working to arrange one, while China’s Foreign Ministry had not confirmed it at the time of publication. But Gates has already made fairly clear what he wants to discuss: whether there should be thresholds for releasing dangerous models, whether AI capabilities that could facilitate biological threats should be monitored, and whether China and the United States can establish some shared rules before competition becomes even more intense. Gates even told Reuters that Xi might be surprised by how strongly he now feels about these issues.

That is interesting because Gates did not sound like this three years ago. In July 2023, he published an essay titled The Risks of AI Are Real but Manageable. His basic argument was that humanity had dealt before with the harmful side effects of new technologies—from automobile accidents and cybercrime to nuclear proliferation and disinformation—and that risks could generally be brought under control through better laws, better technology, and social adaptation. He acknowledged that superintelligence might raise a different class of problems, but when it came to the AI of the day, he still believed that historical experience offered useful guidance.

As recently as January 2026, Gates was still describing himself as an “optimist with footnotes.” AI, he believed, would create enormous wealth and improve medicine, education, agriculture, and scientific research, though job displacement and malicious use had become serious concerns. Seven months later, his language had changed again: “This time really is different.” That may be the most important sentence for understanding why Gates now wants to meet Xi Jinping.

1. What Gates Has Really Changed His Mind About Is Time

Gates has not suddenly become anti-technology. He still believes AI could become “the greatest equalizer in history,” giving a rural doctor access to knowledge once available only in major hospitals, providing African farmers with advice on weather, pests, disease, and soil, and allowing a small business to command research, programming, and analytical capabilities that once required a large company.

What has changed is something else. In the past, Gates believed society would have time to adapt to technology. Now he is beginning to doubt whether social institutions can adjust fast enough to keep pace with the machines.

When personal computers entered the workplace, companies had to buy machines, install software, train employees, and redesign work processes. The transition took more than a decade, in some cases much longer. AI is different. It already lives inside the computers and phones we use. People do not have to learn the language of machines; machines are learning the language of people. More important, if one company discovers that AI can reduce a ten-person job to three people, its competitors will find it difficult to preserve the old staffing model for very long. At that point, what drives change is no longer the moral choice of an individual employer. It is competitive pressure spreading through the market.

This is what Gates now worries about most: technological progress can move exponentially, while social institutions move linearly. Schools may need years to redesign curricula. Tax systems may take years to change. Retraining systems take even longer to build, and social safety nets are often reformed only after a crisis has already arrived. AI models, meanwhile, can improve dramatically in a matter of months.

So Gates’s deepest fear is not “the machines suddenly rebel,” nor is it simply that “AI will take our jobs.” It is a more ordinary, and perhaps more plausible, scenario: AI gradually becomes more reliable, companies gradually hire fewer people, young workers suddenly discover that the first rung of the career ladder has disappeared, and governments begin debating what to do only after unemployment and political anger are already visible.

Recent research offers some support for Gates’s concern, though not for the most pessimistic version of it. An August update from Stanford’s Digital Economy Lab found that among occupations with high exposure to AI, employment among workers aged 22 to 25 had fallen noticeably relative to peers in less exposed occupations, with the gap estimated at 19 percent. At the same time, the researchers stressed that they had not found evidence of broad AI-driven job displacement across the entire U.S. economy. Yale’s Budget Lab has been more cautious still. As of June, it had found no clear evidence that AI was yet reshaping overall U.S. employment, wages, or occupational structure. These findings are not necessarily contradictory. One looks at a vulnerable part of the labor market; the other looks at the whole economy. In other words, mass AI unemployment has not arrived, but some young workers may already be the canaries in the coal mine. Gates is worried about what comes next, not only about what has already happened.

2. The First Thing He Probably Wants to Tell Xi: We Cannot Turn AI Safety into a Game of Chicken

This is probably the most important part of Gates’s proposed trip. He told Reuters that if the United States imposed release restrictions on AI models capable of causing severe harm, China might be willing to adopt similar measures. He has also called for monitoring models that could assist in designing dangerous biological agents or supporting biological attacks, along with the creation of new international mechanisms for AI governance.

The problem is a classic one in game theory. Suppose an American AI company decides that a model is too dangerous and delays its release. Its immediate fear will be: what if a Chinese company releases one anyway? If China restricts a model, it faces the reverse question: will the United States use the opportunity to seize the market? The collectively safest choice may therefore be for everyone to slow down, while the individually safest choice for each player becomes: I cannot afford to slow down first.

That is why arms races are so difficult to stop. Gates is not trying to persuade China to trust the United States. He is trying to identify a few areas in which the two countries can cooperate without requiring a high level of mutual trust.

If a model can substantially increase someone’s ability to create a biological weapon, should it have to pass standardized testing before release? If an AI system demonstrates unexpected autonomous offensive behavior during cybersecurity testing, should companies be required to report it? At what capability threshold should a model face more stringent safety evaluations? If a serious AI accident occurs, should China and the United States have an emergency communication channel?

These are much more practical questions than arguing over whose “AI values” are superior. Nor would such cooperation begin from nothing. China and the United States held their first intergovernmental dialogue on artificial intelligence in Geneva in 2024, discussing risks from advanced AI and global governance. After the U.S.-China presidential meeting in May 2026, the two countries also agreed to launch a new round of government-to-government dialogue on AI. So if Gates does meet Xi in November, he would not be inventing a new agenda in Beijing. He would be trying to add momentum to a political process that already exists but remains extremely fragile.

3. Comparing AI Governance to Nuclear Arms Control Is Only Half Right

Gates often points to nuclear inspections, international aviation rules, and the treaties that protected the ozone layer. He believes future AI governance may have to borrow elements from all three. The comparison is useful, but only up to a point.

Nuclear weapons have several features AI does not. Uranium-enrichment facilities are large. Missile bases are difficult to hide in a pocket. Nuclear tests leave physical traces. AI models are different. They are software. They can be copied, moved across borders, and run by companies, universities, or eventually even individuals. The same model that helps discover a new medicine may also help identify a vulnerability that can be exploited in an attack.

For that reason, the world may find it extremely difficult to create a true “International Atomic Energy Agency for AI.” A more realistic approach would be not to regulate all AI, but to regulate a small number of especially dangerous capabilities. Countries could cooperate on biological-safety testing, establish red lines around cyberattack capabilities, create model-incident reporting systems, and exchange testing methods for the most capable systems. Such arrangements would be less grand, but they might actually work.

This is also where the central difficulty of international AI governance lies. We should not imagine that China and the United States will suddenly trust each other. They do not need to. If both sides can recognize that certain forms of loss of control would harm them both, there is already room for cooperation. The United States and the Soviet Union signed arms-control agreements during the Cold War not because Washington trusted Moscow, but because both understood that nobody could win a nuclear war. AI governance may ultimately follow the same logic.

4. The Second Question Gates Wants to Raise: Who Pays the Price of Greater Efficiency?

The most controversial ideas in Gates’s essay are not about biological weapons. They are two policy proposals that sound surprisingly unlike the traditional Bill Gates. One is something he calls “Human Reserved”—areas of life or work that would remain reserved for human beings. The other is taxing AI tokens and robots.

“Human Reserved” does not mean that humans necessarily perform a task better than AI. It means that even when AI can do something, society may deliberately decide that certain things should still be done by people. Gates came to this idea partly through his father’s experience with Alzheimer’s disease. The caregivers who looked after his father sometimes knew what he needed without a word being spoken. That convinced Gates that some forms of care possess a human value that cannot easily be reduced to technical performance. A machine could, in principle, tell a patient that he has terminal cancer. But society might still decide that such a moment should not be delegated entirely to a machine.

The idea is powerful. Turning it into policy is much harder. What exactly belongs inside a “human reserve”? Doctors? Nurses? Teachers? Judges? Psychotherapists? And if an AI doctor eventually makes far fewer diagnostic errors than an average human doctor, should society still restrict its use merely to preserve human jobs?

Perhaps what deserves protection is not a list of occupations but a set of human responsibilities: who is accountable for a decision, who bears moral responsibility, who has the authority to say no, and who must remain present when a patient, student, or citizen is at his most vulnerable. These may matter more than a crude rule declaring that a certain percentage of work must be performed by humans.

Gates’s other proposal—taxing robots and AI tokens—is not entirely new. He was already talking about a “robot tax” in 2017. What he has now done is extend the logic to generative AI. If a company pays payroll taxes and social insurance when it hires a human being, but can deduct the purchase of a robot as a capital investment, the tax system may effectively encourage the substitution of machines for people.

The problem is real, but taxing tokens directly may not be the best solution. Tokens are essentially an input into AI computation and services. A blanket tax on them could also penalize the very uses of AI society wants to encourage, including medicine, education, and scientific research. Companies could also shift computing activity to lower-tax jurisdictions.

A more sensible direction may be to rethink the balance between the taxation of labor and capital, so that some share of the enormous gains generated by automation flows into unemployment insurance, retraining, and social protection. The real question is not whether a robot should pay taxes. It is this: Who ultimately receives the productivity gains created by machines?

That may be one of the hardest economic questions of the AI age.

5. Gates May Discover That China Faces a Different Problem

When Americans talk about robots, they often begin with job losses. China may see the issue differently. China is dealing simultaneously with population aging, a shrinking working-age population, and intense pressure to remain competitive in manufacturing. From that perspective, robots are not simply a threat to employment. They may also be one of the tools China needs to sustain productivity in the future. Gates himself acknowledges that the costs and benefits of automation and machine-based care can look very different in a society where the labor force is shrinking than in one with a large young population.

That means Gates is unlikely to get very far with Xi if his message is simply, “Robots are advancing too quickly; we need to protect jobs.” The more relevant question for China is different: Where should machines be allowed to move faster, and where must society be given time to catch up?

In factories, replacing dangerous, repetitive, physically punishing work with machines may be an unambiguous gain. But if millions of young people suddenly lose access to their first white-collar jobs, the problem looks very different. AI that helps primary-care doctors make better diagnoses is progress. A child spending six hours a day with a virtual companion that never disagrees with him raises a different set of concerns.

On that latter issue, China has already moved faster than many countries. The Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interactive Services, which took effect on July 15, 2026, prohibit AI services from encouraging emotional dependence and bar providers from offering minors virtual intimate relationships such as virtual companions or virtual relatives. The rules also require special protections for minors and reminders about time spent using such services.

Gates specifically mentioned China in his own essay. That is an easily overlooked detail. In the usual story of U.S.-China AI competition, China appears mainly as America’s technological rival. In Gates’s current framing, China is also becoming a testing ground for the social governance of AI.

6. What Gates Really Wants to Tell Xi May Come Down to One Sentence

Put Gates’s long essay together with his Reuters interview, and I do not think his real message is that China and the United States should stop competing. That would be unrealistic, and Gates knows it. What he seems to mean is something closer to this:

Of course we will continue to compete. But we have to decide what must not become a competition.

Who develops a better cancer drug first? Compete. Who builds a cheaper robot first? Compete. Whose model is smarter? Compete. Who uses AI to raise productivity more effectively? Compete.

But who is first to release an inadequately tested model capable of helping people create biological weapons? That should not be a competition. Who is first to remove every layer of human oversight? That should not be a competition either. Nor should anyone want to win a race to make children heavily dependent on AI companions.

If that reading is right, then Gates’s possible trip to China is about more than a retired technology billionaire visiting the leader of a major power. Behind it lies a much larger question. For the past forty years, U.S.-China technological competition has mainly been about who can run faster. AI is forcing both countries to confront a different question for the first time: Are there places where the person who runs fastest may end up taking everyone else over the cliff with him?

That is why Gates’s change of tone matters. He has not abandoned his old technological optimism. He has simply added a variable that was too often missing from the model: society.

The history of technology is never only the history of inventing machines. It is also the history of inventing institutions that allow human beings to live with those machines. After the automobile came the driver’s license, the traffic light, insurance, speed limits, and rules of the road. After the airplane came air-traffic control. After modern pharmaceuticals came clinical trials and regulatory approval.

AI, too, will eventually need its own traffic lights.

The question is whether, now that the car is accelerating, China and the United States can agree on what a red light means before the crash.

Subscribe now

Li Zhide

A recent case in Florida suddenly brought together artificial intelligence, love, violence, privacy, and public safety. Darren Zhou, 25, continued contacting his former girlfriend by phone, text message, and social media after she ended their relationship of about six months. At the same time, in extended conversations with ChatGPT, he discussed plans to stalk, kidnap, sexually assault, and kill her, and then take his own life. After OpenAI detected troubling content, the conversations were escalated for review, and the company eventually alerted the FBI. The FBI passed the information to authorities in Palm Beach County, who contacted the woman and opened an investigation. Zhou was arrested in May. On August 13, he pleaded guilty to three charges: aggravated stalking, written threats to kill, and unlawful use of a communications device. The judge withheld a formal adjudication of guilt and sentenced him to eight years of probation, with electronic monitoring during the first two years. Judge Scott Suskauer said one important reason he accepted the plea agreement was that the victim herself had agreed to it.

Some headlines reduced the case to something like this: “Man whispers secrets to ChatGPT; AI calls police and gets him arrested.” It is catchy, but not quite accurate. ChatGPT did not hear one dangerous sentence and immediately place a call to the FBI like a robot in a movie. OpenAI has explained publicly that its current process begins with automated systems identifying content that may involve serious violence. Trained reviewers then examine the material in context. Only when a threat is judged to be imminent, credible, and capable of causing serious real-world harm to another person may the matter be reported to law enforcement. In other words, this was not “a robot calling the police.” It was a chain of responsibility: a machine detected signals, human beings assessed the risk, law enforcement investigated, and a court ultimately decided how the case should be handled.

But to reduce the question to whether ChatGPT has the right to alert police would still miss the larger meaning of the case. What makes it worth thinking about is that an old human problem has reappeared in the age of artificial intelligence. There is an old Chinese saying: “When a decision must be made but is not, disorder follows.” Versions of this idea appear in records going back at least to the Records of the Grand Historian and the Book of Han. More than two thousand years have passed. Human relationships have changed. Communications technology has changed. Artificial intelligence has even entered private life. Yet some human difficulties remain almost unchanged: a relationship is over, but one person refuses to accept that it is over; warning signs appear, but the people involved and those around them hope the danger will simply fade away. A problem that might once have ended is allowed to grow into something much larger.

Start with the relationship itself. According to police materials, the woman did not fail to break up with Zhou. Quite the opposite. She had already ended the relationship, and because she was worried about how he might react, she chose to do so by phone. Zhou continued contacting her afterward, even though she repeatedly told him to stop. The material later reviewed by police showed an increasingly clear pattern connecting phone calls, text messages, numbers generated through different apps, social media activity, and violent statements made to ChatGPT. So the idea that “what should have been cut off was not cut off” must not be used to blame the victim. She was already trying to end the relationship. The person refusing to accept the break was the other party.

That distinction matters. In everyday life, people often imagine that a breakup is something two people must somehow complete together, as though the relationship is not really over until both sides agree. That is not true. A romantic relationship requires the consent of two people. Ending one requires the decision of only one. A person may try to win someone back. He may grieve, lose sleep, regret what happened, or remain unable to move on for a long time. But he does not have the right to demand that another person continue loving him. Love exists only through mutual consent. It is not property that one person acquires and then owns forever.

That is why statements such as “If I cannot have her, no one else can” are dangerous for reasons that go beyond anger. They reveal a way of thinking rooted in possession. The other person is no longer seen as someone with an independent will, but as someone who “belongs to me.” Once love changes from “I hope you love me” to “you must belong to me,” love has effectively ended and control has begun. When control develops into stalking, threats, fantasies involving weapons, and rehearsals of violence, the matter is no longer merely a private romantic dispute. It has entered the realm of public safety.

This is also why it is too easy to ask the woman, “If you were so frightened, why didn’t you call the police earlier?” Researchers who study intimate-partner violence have long recognized that separation does not always end danger immediately. In some cases, it can be the period when danger increases. Research collected by the U.S. Department of Justice has found that controlling partners may continue stalking or attacking former partners after separation, while data from the Centers for Disease Control and Prevention show that current or former male intimate partners account for a significant share of perpetrators in homicides of women. To an outsider, “just call the police” is a simple sentence. For someone who knows the other person’s temper, fears provoking him, and may still carry emotional attachment to a relationship that has just ended, the decision can be far more difficult.

The lesson, then, is not that victims should be told, “You should have been more decisive earlier.” The more important lesson is that society must learn to recognize earlier the signs that a boundary has already been crossed. Repeated harassment is not devotion. Changing phone numbers in order to continue unwanted contact is not persistence. Threats are not the language of love. Stalking is not an attempt at reconciliation. Sending pictures of weapons to frighten someone cannot be brushed aside as mere “emotional instability.” Research on stalking has long warned law enforcement against viewing such signs one by one. They need to be understood as parts of a pattern that may be escalating.

This case also shows one new capacity that artificial intelligence may add to risk detection: it can see a sequence over time that an ordinary person may never see. A friend hears one furious remark and may dismiss it as something said in anger. A police officer sees one text message and may have difficulty judging how serious it is. But if a conversational system sees the same person over several weeks repeatedly mention the same target, locations, timing, weapons, stalking, and ideas of harm, it is no longer looking at a single sentence. It is seeing an increasingly coherent trajectory. When OpenAI described its safety procedures in 2026, it made precisely this point: an individual message may not appear dangerous on its own, while patterns that emerge over a longer conversation, or even across conversations, may reveal a more serious level of risk.

This may be the real reason the case deserves a place in the history of artificial intelligence. In the past, many dangerous thoughts remained inside a person’s mind, in a diary, or scattered across fragments of conversation. Today, more and more people use AI as a secretary, adviser, friend, emotional dumping ground, or even a kind of electronic confessional. That creates a question that did not exist in this form before: How private should a conversation between a person and an AI be?

The answer cannot simply be “completely private,” but neither can it be “the platform may inspect whatever it wants.” If someone says, “I hate him so much I could kill him,” and that alone triggers a police report, we will quickly create a suffocating digital society. Many things people say in anger, grief, or fantasy do not amount to an intention to act. If every dark thought becomes a police matter, people may become afraid to speak honestly to therapists, doctors, friends, or even an AI. A safety mechanism could then end up undermining safety itself.

But the other side of the problem cannot be ignored either. If a conversation moves step by step from emotional venting toward a named target, a real location, a schedule, weapons, stalking, and preparations for action, and the platform still insists, “This is private user information and none of our business,” then privacy may become an excuse for doing nothing. OpenAI’s stated approach tries to draw the line here: automated systems identify signals, human reviewers examine them, and the question then becomes whether there is an imminent and credible risk of real-world harm to another person. The boundary is far from perfect, but it gets at the central issue. What requires intervention is not the existence of a bad thought by itself. It is the point at which a bad thought begins moving toward real action.

For that reason, this case should not become an excuse for unlimited expansion of AI surveillance. A platform is not the police, and an algorithm is certainly not a court. Risk-detection systems can make mistakes. Language contains irony, fiction, role-playing, angry venting, and differences of context. A responsible system must preserve human review, set a high threshold for real-world danger, minimize the disclosure of irrelevant information, and leave actual investigation and legal judgment to law enforcement and the courts. AI may raise an alarm. It must not thereby acquire the power to declare someone guilty.

This is another form of knowing when to cut something off in a modern society. What should be interrupted is the chain of escalating risk, not ordinary private expression. What should be cut is the path from fantasy to real violence, not every zone of privacy between human beings and technology. If the threshold for intervention is too loose, genuine danger may be missed. If it is too tight, we may build a society in which no one ever feels unwatched. The hardest problem of governance in the age of artificial intelligence is not choosing either “security” or “freedom.” It is finding, before harm becomes irreversible, a point of intervention supported by evidence, bounded by clear thresholds, subject to human judgment, and open to oversight.

The case of Darren Zhou also raises another question that can easily be obscured by the spectacle of “AI calling the police.” How does a well-educated young man working at a major financial institution reach this point after a breakup? Education, income, and professional ability clearly do not automatically teach a person how to accept rejection. Someone may master complex financial models and make it to Wall Street, yet never learn one of the simplest and hardest lessons in life: another person has the right to leave you. Court materials indicated that his defense described his condition at the time as a serious mental-health crisis. But psychological distress should not become a synonym for violent threats, nor can it erase responsibility for specific conduct. The overwhelming majority of people who experience mental distress do not harm others. What still needs to be discussed are the actions themselves: control, harassment, threats, and repeated violations of another person’s boundaries.

This may be one of the most underestimated problems in education today. We spend years teaching young people how to succeed, but rarely teach them seriously how to fail. We teach them how to compete, but seldom how to withdraw. We teach them how to win love, but rarely how to accept that love has ended. Yet no human life consists only of gaining. Jobs are lost. Investments fail. Friends leave. Parents die. Love ends. A mature person is not someone who always gets what he wants. It is someone who, even when he does not get what he wants, still knows where the line is that he must not cross.

In this sense, “cutting off” is not cruelty. It is an acceptance of reality. If a relationship has ended, let it end. If contact has been rejected, do not begin again from another phone number. If danger signs have already appeared, do not keep soothing yourself with “he is just talking” or “he will calm down in a few days.” Individuals need this ability. Families need it. Companies and governments need it as well. Looking back, many disasters were not preceded by a total absence of warning. Instead, one possible stopping point after another was allowed to pass. Small problems became far more costly because people were unwilling to let go, afraid to stop, or unwilling to admit that an earlier judgment had been wrong.

But the principle of “cutting off when necessary” has another half that is often forgotten: we must know exactly what should be cut off. A person should not be treated as a criminal merely because he says a few vicious things after a breakup. The ability of AI to analyze conversations should not give platforms an unlimited right to monitor people. Nor should one case in which a report may have prevented violence lead us to declare that privacy must henceforth give way unconditionally to security. A mature society does not simply swing a knife at every knot. It learns to recognize which line has actually been crossed before the damage becomes irreversible.

Darren Zhou ultimately did not carry out the most horrifying acts described in the court materials. We cannot prove that he certainly would have committed violence if OpenAI had not reported him. Nor can we prove that his plans would always have remained words. This is precisely what makes risk governance difficult: action must sometimes be considered before the ending is known. What police eventually had was not only language from ChatGPT but also a pattern of continued contact, harassment, and threats in the real world. This was therefore not a case in which one angry sentence caused an arrest. It was a case in which digital traces and real-world behavior gradually converged into a risk pattern that law enforcement had reason to investigate.

More than two thousand years ago, people warned that when a necessary break is not made, disorder follows. The age of artificial intelligence adds a new layer to that old insight. Between people, what must be broken is the claim of possession over a relationship that has already ended. In the face of violent risk, what must be interrupted is the chain that leads from threat to action. Technology platforms must cut off AI assistance that contributes to real-world harm. At the same time, public power must hold another line and refuse to let “security” become a passport to unlimited surveillance.

So what deserves to be remembered about this case is not that “ChatGPT informed on its owner.”

What matters more is that, in a society where more and more human activity leaves a digital trace, we may now have the ability to see certain chains of risk before some tragedies occur. The question is no longer simply whether machines can see them. It is whether human beings have the judgment to decide when to keep watching and when to intervene; when to respect one person’s secrets and when another person’s life must take priority.

The real meaning of “when a break must be made, hesitation breeds chaos” is not impulsiveness. It is a sense of boundaries. It is judgment. And it is a society’s ability to say, before something becomes irreversible:

This goes no further.

Subscribe now

Er Ge (二哥)

There are countless “true stories” online. Many have been edited and reshaped, and some may have begun as advertising scripts. From the material now available, there is no way to know how much of this story—the people involved, the earthquake, the woman’s education, the details of the marriage—is actually true. But authenticity is not the only issue here. Even if the entire story was invented, it is still promoting a set of values. And if the story is true, it turns an extremely complicated private relationship into a public moral example. So the question worth asking is not whether “this woman is foolish,” or whether “this old man is bad.” The real question is: What is this video telling young people?

According to the video, the girl lost both parents in an earthquake when she was eight. A deaf and mute man twenty years older than she was rescued her from the rubble. His face was injured while saving her. He later took her home, raised her for thirteen years, and paid for her education. After she became an adult, the girl remained convinced that she had to repay him. She says it plainly: “Ever since I was little, I told myself that when I grew up, I had to make him the happiest man in the world. I had to marry him.”

That sentence matters. It shows that the logic of marriage constructed by the video is not simply about love. It is this: he saved me, he raised me, he sacrificed so much for me, so I have to marry him. Later comes an even more direct sentence: “I have to give you a child to carry on your family line.” She even folds her determination to have a child, and the risks of childbirth, into the duty of repaying him and completing his life as a man. At that point, this is no longer simply a love story. It becomes a complete chain of obligation: debt of gratitude—debt—marriage—childbearing—sacrifice. And it is precisely this chain of reasoning that deserves examination.

No Blood Relation Does Not Mean There Are No Ethical Boundaries

If two people are not related by blood and are both fully competent adults, an age gap alone is of course no reason to regard their marriage as unlawful. Nor should a modern society deny someone the right to love or marry because that person is disabled, poor, or poorly educated. So the problem is not, “How can a university graduate marry a deaf and mute man?” Criticism framed that way would miss the point. Disability is not a moral failing. Education is not a ranking of human worth. Rural life does not make anyone inferior.

The real question is what kind of relationship had already formed between these two people. When the girl was eight, the man was twenty-eight. He rescued her from the rubble and then took responsibility for her daily care for the next thirteen years. According to the video itself, he even wrote to her: “Little sister, your parents are gone. Why don’t you come with me from now on and let your big brother take care of you?” Whether she called him “big brother” or “father” is not really the point. What matters is the social role he occupied. An eight-year-old orphan facing a twenty-eight-year-old adult who is responsible for her food, clothing, education, safety, and upbringing is clearly not in an ordinary friendship between two equal adults, much less in a romantic relationship that existed from the beginning. It is, first of all, a relationship of caregiving and upbringing.

When an adult raises a child over many years, a very particular psychological structure develops: trust, dependence, gratitude, authority, a sense of safety, and a debt of care that cannot simply be measured in money. This is why human societies have never defined kinship ethics by DNA alone. Why can a family not simply say that because there is “no blood relation,” a father-daughter or sibling-like relationship may later be freely redefined as a romantic one? Because society has to protect the boundaries of caregiving itself. A child must be able to trust that the person raising her is caring for her because she is a child who needs care, not because that care might one day become a form of romantic or marital capital.

That is an important boundary of civilized life. So it would be inaccurate simply to call this case “incest” in the legal sense. But it is entirely reasonable to see ethical risks in the transformation of a quasi-familial or de facto caregiving relationship into a marriage. The issue is not their genes. It is the roles they occupied in each other’s lives for more than twenty years.

If the woman had said, “After I became an adult, I slowly realized that I had fallen in love with him,” that would at least be a narrative of romantic love. But the central sentence in the video is different: “He has done so much for me.... I have to marry him.”

The word “have to” exposes the problem. Love involves choice. A debt demands repayment. Once someone thinks, “Because you raised me, I have to marry you,” marriage is no longer simply an expression of love. It has become, at least in part, a means of repayment. And one of the central principles of modern marital ethics is precisely this: marriage must not become an instrument for paying off a debt.

If someone saved your life, of course you should be grateful. If someone raised you for thirteen years, there is even more reason for gratitude. But why must gratitude become marriage? There are many other ways to repay kindness. You can support someone in old age, pay for medical care, improve their housing, keep them company, help them live their later years with greater security and dignity, support them financially once you begin earning money, or bring them somewhere with better living conditions. Why, among all the possible forms of gratitude, does the video insist on presenting “marrying him” as the highest form of repayment?

“Offering Oneself in Marriage” Is a Deeply Outdated Moral Idea

Behind this story lies a very old idea: the most precious thing a woman can give is herself, followed by her body, her marriage, and her reproductive capacity. The traditional phrase, “I can never repay you for saving my life; I can only offer myself to you in marriage,” rests on exactly this logic. Such stories flourished in eras when women had little property, few professional opportunities, and limited economic independence. A woman had few resources available to exchange, so in the end what remained was marriage and her body.

But a modern society should be moving away from precisely this idea. A woman can have knowledge, a profession, expertise, the ability to earn an income, and the capacity to act in society. She has far more ways to repay kindness than women in the past did. If, after all that, we still tell her that the greatest repayment she can offer is to marry herself off, then what exactly has education changed?

The video’s story about childbirth is even more troubling. The older man worries that because of his age they should not have children. He fears that if he dies early, a child would become a burden on his wife. If this is true, it actually suggests that on this question he was at least thinking about her future. But the woman replies: “I have to give you a child to carry on your family line.” She then says that in order to have this child, “I was willing to risk my own life.” The host immediately responds: “What a great woman.”

The pattern is unmistakable. A woman first repays kindness through marriage, then through childbirth, and finally proves her love by risking her life in giving birth. The greater the sacrifice, the greater the moral worth. The more of her body she gives, the truer the love. This is no longer simply a story about affection between husband and wife. It is an old narrative of feminine virtue: a good woman repays kindness; a good wife gives her husband an heir; a good mother is willing to risk death in childbirth; and through such self-sacrifice, she becomes “great.”

The question is why a modern society should continue treating female self-sacrifice as the most moving proof of moral worth. If a woman wants a child because she herself wants one, there is no problem. But saying, “I have to give you a child to carry on your family line,” is something else. It turns reproduction from a life choice made by two people into a woman’s duty to continue a man’s bloodline. It follows exactly the same logic as “I have to marry him.” The marriage is no longer entirely hers, her body is no longer entirely hers, and reproduction is no longer entirely hers. All become instruments for repaying a debt of gratitude. That is what deserves the strongest criticism in the video.

If the story in the video is substantially true, then the older man did something extraordinary. A twenty-eight-year-old deaf and mute orphan, with limited means of his own, rescued an eight-year-old girl from the rubble and then raised her for thirteen years. That alone is deeply moving. It does not need the later marriage to make it meaningful.

What deserves admiration is the rescue and the years of care. If his purpose in raising the girl was to help her grow up, receive an education, and have a better life, then what would be the best possible outcome of that caregiving relationship? It should be the moment when she finally becomes independent. A truly successful caregiver does not make a child remain indebted forever. He helps the child reach the point where she no longer needs him. That is one of the most admirable things about education and caregiving.

So the best way for the woman to repay him after growing up might be to use the abilities she has acquired to improve both their lives. She could become a doctor. She could earn money. She could help him receive better medical care, improve their housing, provide for him in old age, and make sure he no longer has to depend on street vending to survive. This does not mean she must remain in a city, much less that rural life is somehow unworthy. Adults have every right to choose life in the countryside, just as they have every right to give up a high-paying career.

The real question is this: did she return because she genuinely wanted that life, or because she believed she owed him her life and therefore had to surrender the whole of her future?

The two choices may look exactly the same from the outside, but ethically they are very different. Real freedom is not about whether one lives in a city or the countryside. It is about whether one believes one has the right to choose.

Consent Is Not the End of Ethical Analysis

Some people may say: they are both adults. One wants to marry, the other is willing. What is there to criticize? In law, voluntary consent between adults is of course one of the most important principles. But social ethics cannot stop at the word “consent,” because human desires do not grow in a vacuum. A girl who was rescued at eight by an adult, then raised and supported by him for thirteen years, cannot have exactly the same psychological relationship with him as she would with an ordinary romantic partner she first met at twenty-five.

She herself describes the psychological pressure: “He has done so much for me.” “I have to marry him.” Notice that she does not say, “I choose to marry him.” She says, “I have to.” That sense of obligation may come entirely from within her. No one may have forced her. But an internal sense of duty can also constrain freedom.

A person may sincerely believe: my parents raised me, so I have to live the life they arranged for me; my teacher helped me, so I must never disagree with my teacher; my workplace trained me, so I can never leave; someone saved my life, so the rest of my life belongs to him. All of these beliefs may arise voluntarily. They are still worth questioning.

So an important distinction has to be made: consent does not mean the absence of pressure created by indebtedness. Nor does it mean that every voluntarily chosen value should be held up as a social ideal. “She has the right to make this choice” and “we should applaud this choice” are entirely different statements.

At the same time, the ethical unease surrounding this marriage does not justify the conclusion that “this man rescued her from the beginning because he was waiting to marry her later.” There is no evidence for that.

The video even contains details pointing in the other direction. According to the woman, she was the one who persisted for years, while the older man did not immediately accept her proposal. Later, because of his age, he also suggested that they not have children, fearing that he might die early and leave her burdened with a child. These details at least mean that, without further evidence, we cannot claim that he began raising her at twenty-eight in anticipation of sex and marriage twenty years later. Such an accusation may feel satisfying to make, but it replaces the real ethical problem with an unsupported personal attack.

Nor should his deafness or lack of a middle-school education be used to portray him as an “inferior person.” Disability is not a defect of character, and education is not a qualification for marriage. If the criticism becomes, “How could a graduate of a prestigious university marry an uneducated deaf man?” it simply falls into another form of prejudice.

The real question has never been whether he is good enough for her. It is this: should someone who has raised another person for years accept marriage from that person when the marriage is explicitly offered as repayment?

If an adult once occupied the role of a father, older brother, or guardian, then even if the person he raised later expresses romantic love after reaching adulthood, he has a greater ethical responsibility to proceed cautiously than an ordinary romantic partner would. He knows that her love may be mixed with gratitude, dependence, admiration, a sense of safety, and even the feeling that “I owe you my life.” In such a situation, mature love may not always mean accepting the offer. Sometimes it means helping the other person ask a harder question: Do you love me, or do you feel that you have to repay me?

“Offering Oneself in Marriage” Is a Life Organized Around Debt

Watch the video from beginning to end and its structure is strikingly polished. The first layer is catastrophe: eight years old, an earthquake, both parents dead, two days under the rubble. The second is sacrifice: the man rescues the girl, a steel bar cuts his face, leaving him disfigured. The third is vulnerability: orphaned, deaf and mute, poor, little education. The fourth is reversal: the girl grows up, attends a prestigious medical university, and becomes young and accomplished. The fifth is dramatic contrast: a highly educated young woman, an older deaf and mute man, twenty years between them. The sixth is repayment: “I have to marry him.” The seventh is greater sacrifice: “I have to give you a child to carry on your family line.” “To have this child, I was willing to risk my life.” The eighth is the host’s final moral seal of approval: “What a great woman.”

Videos like this have another old habit: the greater the suffering, the truer the love; the poorer the life, the purer the feeling; the more a woman sacrifices, the greater her moral worth. It is a very old aesthetics of suffering.

But real poverty is not romantic. An older disabled person needs stable income, medical care, accessible services, and support in old age. A young person needs work opportunities, social participation, and the ability to live independently. A family needs housing, education, and health care. If the story ultimately tells viewers that the most moving ending is for a highly educated young woman to give up other possibilities and place herself back into the poverty from which she came, then we should ask a simple question: Why?

Of course, she is free to choose that life. But the media should not automatically equate “giving up the most” with “loving the most.” Hasn’t much of human civilization’s long effort been about allowing people to sacrifice themselves a little less?

What deserves admiration is not: “I am willing to keep suffering with you.” A better outcome would be: “I now have the ability to help us suffer less.”

Strip away all the details, and the video is really promoting a familiar view of life: when someone is good to you, you owe them; the greater the kindness, the greater the debt; the greater the debt, the more of yourself you should sacrifice; and in the end, even your body, your marriage, your reproductive choices, and your entire life may be offered in repayment.

This logic extends far beyond relations between men and women. Parents say: I raised you, so you should live the life I choose for you. A teacher says: I trained you, so how can you oppose me? A workplace says: the organization invested in you, so how can you leave? Some people even treat public services as favors bestowed from above, so that anyone who receives them supposedly loses the right to criticize.

It is all the same way of thinking: turning kindness into a claim of ownership. Healthy gratitude should work in exactly the opposite way. Good parents raise children so that one day those children can become independent. Good teachers help students so that one day the students can surpass them. And a truly selfless act of saving someone’s life should never become ownership over the rest of the rescued person’s life.

The best way to repay kindness is often not to return it to the same person in the same form, but to pass it on. Someone who was once rescued later rescues another person. Someone who received an education later helps others learn. Someone who was treated kindly learns to treat strangers with kindness. That is how gratitude stops being a private debt and becomes a public good.

If children and young people are going to see this video, what they need to hear is not “What a great woman,” nor simply the insult that “there must be something wrong with her head.” They need to understand several more important principles: if someone saved your life, you should be grateful, but your body is not a debt; if someone raised you, you should repay that care, but your life is not collateral; marriage may grow out of love, but it should not become a way to repay the cost of being raised; having children may be a shared desire between two partners, but it should not become a woman’s obligation to “give a man an heir”; sacrifice can sometimes be admirable, but sacrifice itself is not the same thing as virtue.

What young people should be taught is not to “offer themselves in marriage.” The best outcome after receiving help is not to remain forever on one’s knees before a benefactor, but to be able one day to stand on one’s own feet. Seen from this perspective, the most troubling part of the video may not even be the marriage itself. It is the assumption behind the story: that marrying her benefactor is still not enough, that a woman must also bear him a child and risk her life doing so before she can be called “a great woman.” That is a deeply outdated view of life and morality.

The dignity of a modern woman does not lie in proving how much she is willing to sacrifice for others. It lies in possessing a full and independent self: able to love and also to refuse; able to feel gratitude and also to remain independent; able to care for others while still knowing that her own life belongs to her.

Caregiving is caregiving. Gratitude is gratitude. Love is love. Marriage is marriage. They may, of course, meet in the same life, but they should not be used to settle accounts with one another. The civilized way to repay kindness is not to give yourself back to the person who helped you. It is to have been treated well, and therefore to become capable of passing that kindness on. If a society still needs “offering oneself in marriage,” “giving a man an heir,” and “risking one’s life to bear a child” as proof that a woman knows how to repay a debt of gratitude, then perhaps the person who most needs to be re-educated is not only the woman in the video.

Subscribe now

Li Zhide

The claim that “China now has 320 million flexible workers” has recently attracted a great deal of attention online. Some see it as evidence of a new economy taking shape. Food delivery, ride-hailing, livestreaming, e-commerce, self-media, domestic services, freelance design, and independent programming have all created ways to earn a living outside traditional organizations. In the past, many people had to enter a company or public institution to make money. Today, in some occupations, one person with a smartphone can work independently. Others draw the opposite conclusion: if 320 million people do not hold formal jobs, China’s labor market must be in serious trouble.

The figure of “320 million flexible workers in 2026” is not an official statistic already released by China’s National Bureau of Statistics. It is a projection from the 2025 China Blue-Collar Employment Report, released in June 2026 by the China Research Center for New Forms of Employment. The report estimated that China had about 280 million flexible workers in 2025 and projected that the number could reach 320 million in 2026. According to Caixin, the study drew on 28,450 valid survey responses, along with other data and field research. One official benchmark is clearer: around 2021, China already had roughly 200 million people engaged in flexible employment. The category itself is broad. It includes self-employment, part-time work, and new forms of platform-based work. It is not limited to food-delivery riders and ride-hailing drivers.

The increase from 200 million to 280 million took about four years, or roughly 20 million people a year on average. The 320 million figure is still a projection. Definitions may also differ across years, so figures from separate reports cannot simply be stitched together and treated as one consistent official time series. Even so, 320 million is not a number that can be dismissed. The real issue is not whether the correct figure is 280 million, 300 million, or 320 million. The larger change is clear: China now has an enormous population working outside traditional forms of employment.

What does that mean in political-economic terms?

Risk Is Shifting from Firms to Individuals

The first step is to separate concepts that are often mixed together. Flexible employment is not unemployment. Someone who delivers food for eight hours a day and earns an income is employed. A freelance designer who serves ten companies a year is employed. A person who runs a small shop may receive no salary, but still produces goods or services and earns an income. The National Bureau of Statistics has also made clear that both regular and flexible workers count as employed as long as they provide goods or services in the labor market and generate output. It is therefore wrong to describe 320 million flexible workers as 320 million unemployed people.

But the opposite conclusion is also mistaken. Being counted as employed does not mean all employment problems disappear. Economics still has to ask at least four questions: How much do people earn? How stable is that income? What protections do they have? Is there a path for advancement? A programmer earning 20,000 yuan a month while choosing clients and working hours, and a delivery rider working ten hours a day with income fluctuating according to orders, can both be classified as flexible workers. Their economic lives are completely different.

There are really two kinds of flexibility. One is flexibility enjoyed by the worker: the ability to decide when to work, whom to work for, what to charge, whether to reject an order, and when to change clients. The other is flexibility enjoyed by the firm: workers are called in only when needed; when demand falls, the firm does not have to carry the same wage costs; prices may be set by a platform; assignments may be distributed by algorithms; and the worker may still not be treated as a conventional employee. Both are called “flexible employment,” but their economic meaning is very different. The political-economy question is simple: Who is actually becoming more flexible? If workers gain freedom, that is progress. If firms gain flexibility while risks are shifted onto workers, the picture is very different.

Traditional industrial employment had a relatively clear structure. When a company hired a worker, it generally took on a set of fixed costs: wages, pension contributions, medical insurance, work-injury protection, unemployment insurance, workspace, training, and part of the occupational risk. Even when business slowed, many of these costs did not disappear immediately. Platform work and project-based employment have changed this relationship. A food-delivery job exists only when an order appears. If there is no order, a platform usually does not pay a waiting wage. A company needing a logo can hire a freelance designer and end the relationship once the project is complete. A ride-hailing platform can expand or shrink its pool of drivers according to demand without carrying the same long-term wage obligations as a traditional taxi company.

From the perspective of the firm, this can look like an efficiency gain: fixed costs become variable costs. But costs do not simply vanish. They move. Who bears the risk when there are no orders? The worker. Who bears the risk of being unable to work because of illness? The worker. Who pays for the vehicle, phone, computer, or workspace? Often the worker. Who absorbs the shock when income suddenly falls? Again, mainly the individual and the family.

The deepest change created by flexible employment is therefore not just that “the way people work has changed.” It is that the distribution of economic risk has changed. The International Labour Organization and the OECD have both drawn attention to a similar problem in platform work: some supposedly independent workers are technically self-employed but cannot set prices, determine how work is organized, or control the rules governing the service. They do not enjoy the real autonomy of entrepreneurs, yet they carry many of the operating risks and social-protection burdens that firms once absorbed.

That is why the rise of flexible employment should not automatically be treated either as economic progress or as economic collapse. It is better understood as a repricing of the employment relationship. In the past, capital bought a relatively stable block of labor time and accepted part of the risk attached to the worker. Increasingly, firms buy a task, a delivery, or a project. The company becomes lighter. The worker has to manage more of the uncertainty.

The Biggest Macroeconomic Problem Is Uncertainty

The most important macroeconomic effect of flexible employment may not be the level of wages itself, but the uncertainty of income. Someone who earns 8,000 yuan this month and expects roughly the same next month can plan consumption, borrowing, and saving with some confidence. Someone who earns 10,000 yuan this month but has no idea whether income three months later will be 10,000 or 4,000 behaves differently. Even if the two people earn the same average income over the year, the second has a stronger reason to hold back cash.

Economists call this precautionary saving. That does not mean all 320 million flexible workers are afraid to spend. The group is far too diverse for such a claim. A more defensible conclusion is this: if the share of workers with unstable incomes and weak employment protection keeps rising, while social protection fails to adapt, households will have stronger incentives to save against future risks.

This is not merely an online hypothesis. The World Bank’s 2025 China Economic Update argued that stronger social protection for migrant workers, temporary workers, and similar groups could reduce precautionary saving and support household consumption. An IMF study released at the end of 2025 also linked China’s high household savings to weaknesses in social protection and differences in access to public services associated with the household-registration system. Stronger safety nets, it argued, could reduce the need for households to self-insure and thereby raise consumption.

This connects flexible employment to one of China’s larger economic challenges: domestic demand. Income has two basic destinations—consumption or saving. When the future feels secure, households are more willing to spend. When the future feels uncertain, they must build their own insurance. They may need to prepare for medical bills, unemployment, retirement, and children’s education. The more risks individuals must carry themselves, the more a bank account begins to function like a private social-security fund.

This can create an important macroeconomic chain:

employment uncertainty → more precautionary saving → lower propensity to consume → weaker business demand → more cautious investment and hiring.

Flexible employment does not automatically produce this chain. But if social protection fails to keep pace, it can reinforce it.

Most social-security systems built in the industrial era assumed a basic model of employment: one person working steadily for one organization. The employer contributed to pension, medical, unemployment, and work-injury insurance. Wages arrived monthly. Workplaces were relatively fixed. But when more people serve several firms or platforms at once—driving today, selling online tomorrow, working independently next month—a system centered on a single employer naturally begins to show gaps.

By the end of 2024, 70.57 million people in China were participating in the employees’ basic pension system as flexible workers, while 66.159 million were covered by employees’ basic medical insurance. When the Standing Committee of the National People’s Congress reviewed the government’s report on this issue at the end of 2025, it also noted that participation rates in pension and medical insurance among flexible and new-form workers remained relatively low. This is the part of the concern about “who will support 320 million people in retirement?” that deserves serious attention.

Two popular claims, however, need correction. One is that “China’s pension system is simply pay-as-you-go.” That is not accurate. China’s employees’ basic pension system combines social pooling with individual accounts. The pooled component has strong pay-as-you-go features, while individual accounts have an accumulation element. The law is explicit about this structure.

Another claim is that “China’s pension system will first run a deficit in 2029.” That number comes from an older actuarial projection, not a current official forecast for 2026. A 2019 actuarial report from the World Social Security Research Center at the Chinese Academy of Social Sciences projected, under a specific set of assumptions at the time, that the annual balance of the urban enterprise employees’ pension system could turn negative around 2028, with accumulated reserves potentially exhausted around 2035. It was a scenario designed to warn about reform pressures, not a fixed prediction. Since then, China has introduced national pension pooling, gradual retirement-age reform, and other policy changes.

The underlying challenge remains. But it should not be discussed by recycling an old forecast as if it were a fixed date. The better question is: if employment is becoming more mobile, why must pension, medical, unemployment, and work-injury protection remain so tightly tied to a single employer?

A Notable Institutional Experiment

There is already an interesting experiment underway. China’s occupational-injury protection program for workers in new forms of employment represents an important institutional innovation.

Traditional work-injury insurance is based on a conventional employment relationship: a company has employees, and the company pays contributions. But the legal relationship between food-delivery riders, ride-hailing drivers, and digital platforms often does not fit the traditional model. The pilot program addressed this problem in a simple way: instead of focusing only on the labor contract, it focuses on the order.

When a worker completes an order, the platform pays into an occupational-injury protection scheme. By July 2026, the program had expanded nationwide, using methods such as daily enrollment, per-order contributions, and monthly payment. By the end of June 2026, cumulative enrollment had reached 29.902 million people. By the end of July, it had risen to 34.588 million.

The significance of this experiment goes well beyond work-injury insurance. It demonstrates that social protection can gradually shift from “following the employer” to “following the person and the work.” If someone works for three platforms, all three could contribute according to the actual labor performed. A person might deliver food today and drive passengers tomorrow, while the protection account remains attached to the individual.

If technology and administration allow, pension protection, occupational-injury coverage, and perhaps other benefits could increasingly become portable in this way.

This may make more sense than simply requiring every flexible worker to bear the full cost of social insurance alone. If companies benefit from lower fixed costs through flexible labor, they should also bear part of the social costs created by that employment model. In political-economy terms, this is a question of internalizing costs.

At a deeper level, the growth of flexible employment also changes the bargaining relationship between firms and workers. In a traditional company, an employee at least knows who the employer is. If wages are too low, there is a company to confront. If working hours become excessive, there is an identifiable employer. If an accident occurs, the responsible entity is relatively clear.

Platform work makes this relationship more complex. Prices may be set by an algorithm. Rules may be set by the platform. Workers may be managed through an outsourcing company. When a dispute occurs, the platform may say no formal labor relationship exists, while the contractor may claim it is only a service provider. The worker ends up facing an organizational structure that is difficult to see.

That is why the simple question “Is there a labor contract?” is no longer enough to determine whether a labor relationship is fair. The more important questions are: Who sets the price? Who assigns the task? Who can punish whom? Who controls the data? Who bears the losses? Who can walk away without paying a huge cost?

If a supposedly “independent worker” cannot set prices, cannot directly reach clients, depends heavily on one platform for income, and must accept that platform’s ratings, assignments, and penalties, then the worker may not be economically independent in any meaningful sense.

Recent ILO work on platform labor increasingly discusses worker classification, social protection, algorithmic transparency, and collective bargaining together. These are also issues China will need to continue addressing. “Flexibility” should not become a legal device for avoiding labor responsibilities. Workers who are genuinely independent should have freedom. Workers who are economically subordinate should receive protections that reflect the degree of that dependence.

Political Economy Is Ultimately About a New Social Contract

So what does 320 million really mean?

It does not mean that “China already has 320 million unemployed people.” Nor does it simply mean that “the new economy has created 320 million free professionals.” What it really tells us is this: the labor market has changed, while some of the institutions supporting that labor market still retain the shape of the industrial era.

The old social contract was relatively clear. Firms provided jobs. Workers provided labor. Firms absorbed part of the risk. The state built social protection around stable employment relationships.

The next social contract has to look different:

work can be flexible, but protection cannot disappear; labor can be mobile, but rights must be mobile as well.

That means social insurance should increasingly follow the worker rather than a single employer. Platforms that actually organize and control labor should bear social-protection costs proportionate to that control. Pension and medical contributions should allow more flexible schedules and contribution bases, instead of forcing workers with volatile income to carry the same cash-flow burden as workers receiving stable salaries. Employment records, occupational-injury protection, and pension rights should remain continuous when workers move across platforms. When algorithms determine income, assignments, and penalties, workers should have basic rights to information and appeal.

There is also a more basic need: better data.

It is no longer enough simply to publish a headline number showing how many people are engaged in flexible employment. Policymakers should also know how many chose it voluntarily, how many entered it because they could not find stable work, how many hours they work each week, what their median income is, how volatile that income is, how many have pension, medical, and occupational-injury protection, and how many would prefer to move into stable employment.

Only with this information can we determine whether a country’s flexible employment represents greater freedom or greater insecurity.

The number 320 million easily produces two emotional reactions. One is panic: “Half the country no longer has a real job.” The other is optimism: “China has completely upgraded its employment model.” Both are too simple. A serious economic judgment cannot stop at counting how many people are “employed.” It also has to examine the income, stability, productivity, protection, and bargaining power associated with those jobs.

Political economy asks one more question:

When economic transformation creates both gains and risks, who receives the gains and who carries the risks?

If platforms gain efficiency, consumers gain convenience, firms lower costs, and workers receive only uncertainty, such a model of flexible employment will be difficult to sustain. If workers gain genuine freedom to choose how they work while also receiving pension, medical, occupational-injury protection, and opportunities for advancement, flexible employment can become a healthy and effective part of a modern economy.

China’s real task, therefore, is not to force 320 million people back into traditional organizations. That era is not coming back. The real task is to build institutions suited to a new world of work: allow firms to remain flexible while giving workers security; allow technology to raise efficiency while ensuring that workers share in productivity gains; allow people to change jobs, cities, and platforms without leaving their social protection behind.

Only then will “flexibility” truly mean freedom. Otherwise, it simply means turning uncertainty once carried by firms into uncertainty carried by workers. And an economy in which hundreds of millions of people are unsure of what they will earn tomorrow will eventually face more than an employment problem. It will also face problems of consumption, retirement, income distribution, and, ultimately, the way society itself distributes risk.That is the real political-economic meaning of “320 million flexible workers.”

Subscribe now

Li Zhide

On August 17, 2026, the Pentagon suddenly pulled 30 universities into what amounted to a sweeping institutional audit, giving them only two weeks to prepare. Harvard, MIT, UC Berkeley, Duke, Cornell, Johns Hopkins, New York University, and other major institutions were all on the list. They were ordered to review, by August 31, their academic, financial, and research ties with foreign institutions, with particular attention to entities in China, Russia, and Iran that the U.S. government considers national security risks, as well as certain organizations connected to former Confucius Institutes. The universities were also required to examine whether sensitive research or export-controlled technologies might have been exposed, propose corrective measures, and possibly terminate relationships deemed especially risky. Schools that fail to comply could jeopardize their eligibility for future federal research funding. The Pentagon did not publicly explain what specific project or relationship had caused each of the 30 universities to be placed on the list.

A day later, Representative John Moolenaar, chairman of the House Select Committee on the Chinese Communist Party, publicly backed the move. His argument was straightforward: if universities accept American taxpayers’ money, they should not be able to receive federal research funds while collaborating with foreign institutions that the U.S. government has already identified as security risks. On the surface, this looks like another round of research-security inspections. But the story is larger than that. Viewed against nearly a decade of U.S. technology policy, the government is redefining what a university is. Universities were once understood mainly as open systems for producing knowledge. Increasingly, they are being treated as part of the national machinery of technological competition. The old concern was whether an individual professor had secretly taken foreign money. The new concern is whether an entire university has become a channel through which technology, talent, data, and knowledge can flow to strategic competitors. That is the deeper change behind the question of why the U.S. government is now “going after” universities.

American Universities Have Long Been More Than Universities

To understand what is happening, it helps to begin with a fact that is easy to overlook: a large share of America’s most important basic research is funded by the government and carried out at universities. In fiscal year 2023, U.S. universities spent $108.8 billion on research and development, of which the federal government provided $59.7 billion, or 55 percent. In fiscal year 2024, total university R&D spending rose further, to $117.7 billion. American universities are therefore not simply institutions that collect tuition and teach undergraduates. They are also among the most important laboratories at the upstream end of the American innovation system.

This arrangement has deep historical roots. In 1945, Vannevar Bush submitted Science, the Endless Frontier to the president, arguing that the federal government should provide sustained support for basic research in universities. After the National Science Foundation was established in 1950, a broad postwar bargain gradually took shape: the government would provide money, scientists would conduct research, and universities would retain substantial academic autonomy. The underlying belief was that basic research did not need to think every day about making airplanes, missiles, or profits. Give talented people enough freedom to explore, and the knowledge they produced would eventually feed into medicine, industry, defense, and economic growth.

The results of that system are well known. The problem lies in the very source of its success.

In the middle of the twentieth century, there was often a relatively long distance between basic science and military technology. In the age of artificial intelligence, quantum computing, biotechnology, autonomous systems, advanced materials, and semiconductors, that distance is shrinking. An apparently open academic paper, a dataset, an algorithm, or the technical knowledge carried by a visiting scholar can move quickly into industry and, potentially, into military systems. An academic order that became powerful because it was open now faces an uncomfortable paradox: the more freely the United States places its best knowledge inside open universities, the stronger its capacity for innovation becomes; yet as competition with China has intensified, that same openness is increasingly being viewed by national security agencies as a vulnerability. This is not a problem that can be solved simply by declaring openness good or security good.

It would also be a mistake to explain the current campaign simply as the Trump administration attacking universities. That would erase a much longer history. American concern about research ties with China has been building for nearly a decade. In 2018, during Donald Trump’s first term, the Justice Department launched the China Initiative, focusing on economic espionage, theft of trade secrets, foreign talent programs, and researchers who failed to disclose relationships with Chinese institutions. At the time, the Justice Department said that roughly 80 percent of economic espionage prosecutions involved conduct that might benefit China, while about 60 percent of trade-secret theft cases had some connection to China.

But this approach soon created another problem. Some scholars who were investigated were never shown to have committed espionage. Universities and Asian American organizations began asking whether the government was gradually treating a “China connection” as evidence of a “China threat.” In 2022, the Biden administration’s Justice Department ended the use of the name “China Initiative.” Matthew Olsen, then assistant attorney general for national security, explained the problem clearly: threats posed by the Chinese government were real, but putting a wide range of cases under a program explicitly named after China had created the impression that the Justice Department was applying a different standard to Chinese and Chinese American researchers. It could also produce a chilling effect across the scientific community. The Justice Department did not stop pursuing espionage or research-security cases. It placed them instead under the broader Strategy for Countering Nation-State Threats.

So from Trump’s first term, through the Biden administration, and into Trump’s second term, what endured was not a program name but a judgment: international scientific collaboration at universities had become a national security issue. What changed was the method. Earlier enforcement concentrated on individuals: Did this professor hide foreign income? Did he participate in a foreign talent program? Did she transfer technology? The new approach increasingly examines institutions: Who does the university collaborate with? Who funds the work? Who appears as co-author? Which laboratories are jointly operated? Who can access the data? Where do visiting scholars come from? Who jointly supervises students?

That is a significant shift.

The Securing Innovation and Research from Adversaries Act, or SIRA, introduced by Moolenaar in May, shows just how far this shift could go. It is still a bill before Congress, not a government-wide law already in force. Government records show that H.R. 8901 was introduced on May 19, 2026, and referred to the House Committee on Science, Space, and Technology. But the bill’s definition of “research collaboration” reveals how broad Washington’s proposed net has become.

The bill would prohibit people and institutions receiving federal research funding from using those funds to conduct research collaborations with foreign entities on U.S. government restricted lists, or with individuals associated with those entities. More important, “collaboration” would no longer mean merely a formal agreement between two universities. It would include joint research, co-authored publications, data exchanges, shared software or research infrastructure, joint laboratories, visiting scholars, personnel exchanges, and joint supervision of students. In other words, the government is moving beyond the question of whether money is flowing abroad. It is beginning to regulate the relationships through which knowledge itself moves.

Nor is this simply a blacklist of Chinese universities. The legislation draws on multiple government lists, including the Commerce Department’s Entity List and Military End User List, Treasury sanctions lists, the Defense Department’s Chinese Military Companies List, lists involving military-civil fusion, telecommunications security, forced labor, and biotechnology concerns. The bill also contains a waiver mechanism. If a collaboration is necessary for U.S. national security, or serves a clearly defined scientific, public health, or national security purpose that cannot reasonably be achieved in another way, the responsible federal agency could approve it case by case, provided that Congress is informed of the justification and the measures taken to reduce risk.

The underlying logic is now clear: the United States is moving from disclosure of foreign relationships toward restrictions on certain foreign relationships. The older principle was: you may collaborate, but you must tell the government. The emerging principle is closer to this: with some partners, even full disclosure may no longer make collaboration acceptable.

That is an important turn in American research policy.

Why Is Washington Acting So Aggressively Now?

One reason, of course, is China.

In January 2026, the Government Accountability Office released a report examining five major federal research-funding agencies: the Department of Defense, Department of Energy, NASA, National Institutes of Health, and National Science Foundation. NIH data showed that between 2017 and 2024 it received 657 allegations involving “undue foreign influence,” of which 570 involved China. Those numbers do not tell us whether every allegation was substantiated, but they help explain why U.S. security agencies continue to place China at the center of research-security policy.

China is also no longer the country it was twenty or thirty years ago, when it was mainly learning advanced technology from the United States. Semiconductors, artificial intelligence, quantum science, biotechnology, renewable energy, drones, telecommunications, and advanced manufacturing have all become central arenas of strategic competition between the two countries. Many of these technologies are inherently dual-use. An algorithm developed in a university laboratory may one day control industrial robots and, not long afterward, autonomous military systems. Security agencies operating in this environment can no longer comfortably rely on the old assumption that civilian research belongs to universities while military research belongs to the military.

There is another reason: the federal government has developed a clear “trust deficit” toward universities.

The Department of Education reported this year that American universities disclosed more than 8,300 foreign gifts and contracts in 2025, worth over $5.2 billion, of which more than $528 million came from China. An important distinction is necessary here. Foreign funding is not in itself illegal, and it is certainly not the same thing as espionage. Section 117 of the Higher Education Act requires disclosure above specified thresholds; it does not ban universities from accepting all foreign funds.

The problem is that federal officials have found cases in which universities reported funding late, failed to report it, or did not make relationships sufficiently transparent. That has gradually changed the policy logic. If universities cannot reliably manage foreign relationships themselves, Washington will increasingly decide which relationships are acceptable and which are not.

On August 13, just days before the Pentagon’s review of the 30 universities began, two House committees released an investigative report on Harvard. The report accused the university of maintaining relationships with institutions that the U.S. government considers connected to China’s military or implicated in human rights concerns, and it called on Congress to pass the SIRA legislation. These were congressional findings and allegations, not judicial rulings. Still, only days later, Harvard appeared on the Pentagon’s list of 30 universities.

It is difficult to regard the timing as entirely accidental.

But a line has to be drawn here.

The Pentagon has not publicly explained why each of the 30 universities was selected. Officials gave the media the list of institutions, but not a university-by-university account of the evidence. The American Council on Education has therefore objected that the list and the language surrounding it may create the impression that the institutions have already been found to have engaged in improper conduct, even though many universities have spent years adjusting their programs to meet federal research-security requirements, and the higher-education sector itself participated in shaping some of the early security rules.

That criticism is reasonable.

An audit is not a conviction. A relationship with a Chinese institution does not automatically amount to technology theft. A paper co-authored by Chinese and American researchers does not automatically mean that the work assists the People’s Liberation Army. Serious risk assessment has to examine the nature of the research, the partner institution, the identity of the participants, financial relationships, the sensitivity of the technology, and its potential end use.

If “anything connected to China” itself becomes the measure of risk, the United States could easily return to the mistakes exposed by the China Initiative.

The January GAO report is especially interesting because it supports two apparently contradictory conclusions at the same time. On one side, a large number of foreign-influence allegations did involve China. On the other, the GAO found that none of the five major federal research agencies had properly evaluated whether their research-security programs were doing enough to prevent discrimination based on race or national origin. At the Defense Department, some transparency measures, anti-discrimination training, and leadership commitments remained only partly implemented or not implemented at all.

The threat is real. The risk of wrongful suspicion is also real. Recognizing one does not require denying the other.

What America Really Fears Is Losing the Future

If this issue is reduced to “the United States no longer wants universities to work with China,” it is being understood too narrowly. What America is really competing over is the lead time on future technologies.

Basic research has an awkward quality: it is often impossible to know today which discovery will matter most ten years from now. When lasers first appeared, few people understood how deeply they would enter telecommunications, medicine, manufacturing, and weapons systems. The early internet was not designed with e-commerce, short-form video, or today’s artificial intelligence in mind.

This creates a natural anxiety inside national security institutions. By the time a technology has clearly demonstrated strategic value, it may already be too late to protect it. American policy is therefore moving toward pre-emptive protection.

That approach is understandable. It also creates a major side effect.

Why has the United States been able to attract so many of the world’s best scientists for so long? Salaries matter. Laboratories matter. Capital matters. But one factor is often overlooked: openness.

The GAO itself acknowledged this year that America’s rise as a global scientific power depended in important ways on international talent and international collaboration. Research-security work by the National Academies has also warned that initiatives such as the China Initiative made some foreign researchers feel that the American research environment had become less welcoming, even though open scientific exchange is itself one of the sources of American technological strength.

The United States therefore faces a genuine security dilemma. If it keeps the door too open, technology may flow outward in ways that create real risks. If it closes the door too tightly, talented people may stop coming.

And talent itself is technology.

That is why research security cannot simply copy the security system of a military base. A university that treats every foreign researcher as a potential suspect might become extremely secure. It may also stop being a world-class university.

There is another dimension that cannot be ignored: domestic American politics.

The current campaign cannot be completely separated from the Trump administration’s broader conflict with elite universities. During Trump’s second term, the federal government has clashed repeatedly with Harvard and other institutions over federal funding, admissions, campus governance, and related issues. Coverage of the 30-university audit has therefore also placed it in the wider context of the administration’s continuing investigations of Harvard.

Two simple explanations are consequently inadequate.

One says that Trump is merely using “China security” as an excuse to punish liberal universities he dislikes. The other says that the campaign is nothing more than politically neutral national security enforcement.

The reality lies between them.

Research-security concerns have survived both Trump and Biden and have had durable congressional support. They are not a slogan invented in 2026. But the second Trump administration is clearly more willing to use federal funding as leverage, turning what had often been gradual and fragmented security requirements into institutional pressure with short deadlines and explicit consequences.

In other words, the underlying problem has a genuine national security basis, while the way it is being enforced is inevitably shaped by the current political environment.

That distinction matters if we want to understand what is actually happening in the United States.

No one yet knows how many projects at the 30 universities will eventually be terminated. No one knows whether SIRA will become law. But one change has already taken place: the bargain between American universities and the federal government that lasted for more than seventy years is being rewritten.

The old model looked roughly like this: government provided money; universities produced knowledge; the scientific community largely determined research agendas; results were made public whenever possible; talented people came to the United States from around the world.

The new model increasingly looks like this: government still provides money, but it also decides which foreign institutions may not enter federally funded research networks. Universities are expected to examine funding, personnel, publications, datasets, software, joint laboratories, and student-supervision relationships. International collaboration is becoming subject to national-security screening.

This does not mean that American universities have lost their freedom.

It means that the United States is moving from an era of open science to an era of selective openness.

That is also what China should pay close attention to.

This is not simply a story of “America beginning to exclude China.” The United States is gradually placing universities, semiconductors, artificial intelligence, investment, trade, data, and talent on the same national security map. In the past, U.S.-China competition was concentrated mainly in tariffs, corporations, and defense industries. It has now moved into laboratories, lists of co-authors, and relationships between doctoral students and supervisors.

Competition is moving upstream, toward the production of knowledge itself.

For the United States, the central test is whether it can cut off genuinely dangerous collaborations without destroying the openness that helped make it a scientific superpower. If security policy becomes nationality screening, if restricted-entity lists expand without limit, and if every Chinese scholar becomes a presumptive risk, the United States may protect some technologies while damaging its own innovative capacity.

Universities, however, cannot continue to treat “academic freedom” as freedom from responsibility for the sources of their funding and the institutions with which they collaborate. Accepting federal money means accepting a legitimate demand from taxpayers for transparency. Research with obvious dual-use implications cannot pretend that international politics does not exist.

So the real boundary in this debate is not whether research security is necessary. It is how precisely security should be defined and applied.

Should the standard be conduct or identity? Should restrictions target specific institutions or entire countries? Should the government require disclosure, or impose outright bans? When problems are found, should universities be allowed to correct them, or should funding be terminated immediately? Do universities have a right to appeal? Who creates the restricted lists, and how does an institution get removed from one?

The answers will determine whether the United States is building a more mature system of research security or drawing universities into an ever-expanding process of political securitization.

In 1945, America believed science was an “endless frontier.” More than eighty years later, it has not abandoned that belief. But it is now asking, with an intensity it rarely did before:

To whom should the gates of that frontier be open?

Once the Pentagon begins answering that question, the relationship between the American university and the American state is no longer what it used to be.

Subscribe now

Li Zhide

The story of Talatan is easy to tell as a “miracle of Chinese infrastructure.” Solar panels were installed across a barren stretch of high plateau. Grass slowly began to grow in what had once been desert. Then the grass grew too tall and started interfering with power generation. Sheep were brought in. They ate the weeds, reduced the cost of clearing vegetation, and at the same time created additional income for local herders. In the end, several things that had once seemed unrelated—power generation, desert control, and sheep grazing—became part of the same system. It is a remarkable story, and one worth thinking about.

What deserves closer study, however, is not the “miracle” itself, but the pattern of thought and action behind it. The most important lesson of Talatan is not that everything was known in advance. It is that knowledge with real vitality is often not designed once and for all in an office. It emerges through action, is corrected by feedback, and gradually takes shape through setbacks, surprises, and unintended consequences. On the surface, this is a story about solar power. At a deeper level, it touches on an old philosophical question: what is the relationship between knowing and doing? And when that question is enlarged from the cultivation of an individual to the workings of a country, it becomes more important still: How does a nation learn?

I

We should begin by bringing Talatan back from legend to fact. Talatan lies around Gonghe County in Hainan Tibetan Autonomous Prefecture, Qinghai Province, at an average elevation of nearly 3,000 meters. The area receives long hours of sunshine, suffers from severe desertification, and contains large expanses of relatively flat land. For these reasons, it came early into consideration for large-scale solar development. Publicly available information shows that megawatt-scale photovoltaic projects in the area began connecting to the grid around 2011. After 2012, Hainan Prefecture accelerated the construction of large renewable-energy bases, and the photovoltaic park continued to expand.

By 2026, according to figures released by China’s National Forestry and Grassland Administration, the Talatan photovoltaic industrial park had a planned area of 609 square kilometers, of which about 420 square kilometers had already been developed. In 2022, a photovoltaic complex in Hainan Prefecture, with an installed capacity of 8,430 megawatts, was certified by Guinness World Records as the solar power facility with the world’s largest installed capacity. But the more interesting changes were taking place on the ground. Large fields of solar panels altered the original balance of heat, moisture, and wind at the surface. Monitoring data from the park showed that, in some photovoltaic areas, near-ground wind speeds fell by about 50 percent, evaporation of soil moisture declined by roughly 30 percent, and vegetation coverage rose to around 80 percent. The shade and wind protection provided by the panels, together with water used to wash the photovoltaic modules, the artificial sowing of grass seed, and subsequent ecological restoration measures, jointly changed the local habitat.

Academic research offers a more cautious form of support than promotional accounts do. A 2024 study published in Scientific Reports, examining a large desert photovoltaic development zone in Gonghe County, found that solar development had generally positive effects on the local microclimate, the physical and chemical properties of the soil, and plant and microbial communities. At the same time, the study warned against imagining that the photovoltaic zone had therefore become an ecological paradise. Some ecological indicators remained weak, there were substantial differences from one area to another, and the long-term effects still require continued monitoring. That distinction matters. Talatan is not a fairy tale in which solar panels are installed and desert automatically turns into grassland. The real world is more complicated. The panels altered the environment, grass planting helped stabilize the sand, operations and maintenance introduced additional moisture, and management practices were repeatedly adjusted. Only through the interaction of these factors did the landscape we see today emerge.

And the new outcome created new problems. Once the grass began growing, difficulties appeared that had not existed before. Vegetation that grew too high could interfere with some photovoltaic equipment, while dried grass increased the risk of fire. A renewable-energy project had suddenly encountered a livestock-management problem. Around 2015, the park began cooperating with nearby villages, and an initial flock of roughly 600 sheep was introduced into the photovoltaic area. As the sheep grazed, the need for manual weed removal declined. Their manure returned nutrients to the soil and became part of a new ecological cycle. Some projects later adjusted the height and design of photovoltaic supports to reduce conflicts between grazing animals and power equipment. By 2026, Gonghe and Xinghai counties had developed 32 photovoltaic ecological pastures and 56 grazing sites, with roughly 20,600 “solar sheep” being raised across 18 villages.

By this point, the project had moved far beyond the original idea of “building a power station.” The first objective was electricity generation. Later, people discovered that the project could improve the local ecology. Ecological improvement produced grass, and grass then became a new management cost. Sheep helped solve the grass problem, while the sheep themselves became economic assets for local herders. Once the herders entered the system, new rules had to be created for companies, village collectives, grassland management, livestock raising, fire prevention, and ecological protection. An engineering project had become a system. The real value of Talatan lies in that transformation.

II

In traditional Chinese thought, Wang Yangming famously said, “Knowledge is the beginning of action; action is the completion of knowledge.” The line is often reduced to a simple lesson: once you know something, you should act on it. But Wang’s meaning was more demanding than that. What he opposed was the separation of knowing and doing into two completely distinct stages—as if one could first understand the world in full and only then begin to act. For him, genuine knowledge naturally enters into action, while “knowledge” that has never been tested in action may be little more than an idea in the mind.

Seen from this perspective, Talatan becomes especially interesting. No one sitting in a conference room more than a decade ago could have calculated everything that would later happen: how the panels would change wind speed, how they would affect soil moisture, how high the grass would grow, whether that grass would interfere with equipment, how many sheep should eventually be introduced, how the animals could coexist with photovoltaic installations, how village collectives should participate, how herders might benefit, and how overgrazing could be avoided. If action had been postponed until every one of these questions had been answered with certainty, many things would never have begun. But the opposite approach—doing things blindly, without research or restraint—would have been just as dangerous.

The real unity of knowledge and action is therefore not a matter of saying, “Just be bold and do it.” It is a cycle. Existing knowledge gives people enough confidence to begin, action produces new information that did not exist before, that information changes what people think they know, and revised knowledge then changes the next round of action. Knowledge enters action, while action produces new knowledge. Only then do knowing and doing form a complete relationship. From the standpoint of modern epistemology, this is close to the basic spirit of experimental science. The purpose of an experiment is not to prove that the researcher is always right. It is to give reality the chance to tell the researcher where he or she is wrong. The grass at Talatan was one of reality’s answers. The sheep were another.

But moving from the individual unity of knowledge and action to a national version of it requires crossing an important threshold: a country is not a person. A country does not possess a single brain, nor does it have anything resembling an individual conscience concentrated in one organ. Government consists of different departments, regions, companies, experts, local officials, and ordinary citizens. Each possesses different information, and each has different interests.

For that reason, the “national unity of knowledge and action” cannot mean that a leader reaches a conclusion and everyone below simply acts in unison. That can produce the opposite result—a separation of knowledge from action. Senior policymakers possess macro-level knowledge. Engineers possess technical knowledge. Local governments understand local conditions. Herders know when the grass grows and what sheep prefer to eat. Maintenance workers know which kinds of vegetation are most likely to obstruct equipment. Knowledge is dispersed among many people. The true cognitive capacity of a modern state is therefore not measured by how much the people at the top know. It is measured by whether knowledge scattered throughout society can be discovered, transmitted, tested, and eventually incorporated into decision-making.

III

This is the deeper meaning of Talatan. When the grass became too tall, the first people to notice were not policymakers in Beijing. They were people on the site. Nor did an abstract theory first reveal that sheep could clear the grass. Practice did. But if a discovery made in practice remains merely something “one herder knows” or “one power station knows,” it is still only local experience. The crucial step is turning experience into rules.

Talatan today is no longer a place where anyone who wishes can simply bring sheep into the photovoltaic park. Local authorities have gradually developed arrangements under which village economic cooperatives sign agreements with photovoltaic companies, stocking levels are determined according to the amount of available grass, the size of individual grazing sites is controlled, rotational grazing is used, traceability systems are introduced, and grazing intensity is adjusted to the carrying capacity of the grassland. This may sound less dramatic than “the first flock of sheep walking under the solar panels,” but it matters much more. An accidental discovery becomes genuine social knowledge only when it is turned into a stable institution.

This also helps explain why many countries have no shortage of intelligent people and yet continue making serious mistakes for long periods of time. National failure often does not arise because no one knows what is wrong. It arises because what is known cannot enter action, while the information generated by action cannot return to the knowledge system. Experts raise a problem, but no one listens. Local officials discover a mistake, but are afraid to report it. Grassroots units, under pressure to meet targets, report only good news. When a policy fails, no one wishes to admit it. When one case succeeds, it is copied nationwide before the reasons for its success have even been tested. The cycle between knowledge and action then breaks down. The number of documents grows, while the amount of real knowledge may actually shrink.

There is therefore a frequently overlooked measure of state capacity: Can a country learn from what it has already done? One feature of China’s reform history that has long attracted the attention of scholars inside and outside the country is the use of local experiments, comparison, feedback, and gradual expansion. The value of this process does not lie in some assumption that “local governments are always right.” It lies in allowing certain policies to operate first within a limited scope, generating information through actual results, and then deciding whether they should be expanded, modified, or stopped.

This is the kind of national philosophy Talatan can illuminate. The national unity of knowledge and action does not mean “once we know the answer, we must carry it out firmly.” It means building a system of action that is capable of continuous learning. Such a system must possess several abilities. It must be willing to act, because otherwise no new knowledge will be produced. It must allow reality to contradict earlier assumptions, because otherwise action becomes little more than a ritual for proving official documents correct. It must allow bad news from the grassroots to travel upward, because otherwise senior decision-makers will receive only a polished version of reality. And it must know when to expand a successful experiment and when to stop one, because local success does not automatically amount to a universal rule. That last point is especially important.

IV

Talatan should not be treated as a “Chinese answer” that can simply be photocopied indefinitely. This is the most important counterargument the case itself demands. Gonghe County has its own particular combination of high altitude, intense sunlight, dryness, strong winds, soil conditions, grass species, rainfall patterns, pastoral traditions, and population density. In this setting, reducing evaporation and wind speed through photovoltaic installations may help vegetation recover. Somewhere else, the results may be different. In wetter regions, additional shade may alter plant communities in other ways. In ecologically sensitive areas, construction on a large scale may itself cause new damage. In places under extreme water stress, the water used to clean photovoltaic modules must be included in the ecological calculation. Too many sheep could degrade the grassland again. Even differences in panel height, spacing, or tracking technology can affect soil temperature, moisture, and plant composition.

So what deserves to be copied is not the rule that “sheep should always be raised beneath solar panels,” much less the idea that every desert in China should be covered in a replica of Talatan. What deserves to be copied is the method of solving problems. Understand local conditions first, then build. Monitor continuously and identify side effects. Allow adjustment. Bring different departments into the same process. Give local people a stake in the outcome. Then rewrite the rules in response to the new ecological carrying capacity. What can be reproduced from Talatan is its learning mechanism, not Talatan itself.

That distinction matters. A country with weak learning capacity tends to search for standard answers. A more mature country values methods for solving problems.

Take the argument one step further and Talatan reveals another fact that modern societies often forget: many of the most important innovations are not breakthroughs in a single technology. They come from rearranging relationships that were previously kept apart. Photovoltaic technology was not invented at Talatan. Sheep certainly were not. Nor was grass planting for desert control. What was new at Talatan was the way energy, land, water, grass, sheep, herders, companies, the power grid, and local government were connected.

A piece of land that once had only one use can now perform several functions at once: generating electricity, stabilizing sand, restoring vegetation, and supporting controlled grazing. What a power company sees as “weeds” can be feed from a herder’s point of view. If the two systems remain separate, the company pays to remove grass while the herder pays to find feed. Once the systems are connected, one side’s cost can become the other side’s resource.

This is a classic form of systems innovation. Many social problems persist not because people cannot find better individual technologies, but because every department optimizes only its own small piece of the problem. The energy department watches kilowatt-hours. Environmental agencies watch vegetation. Agricultural authorities watch livestock. Finance departments watch expenditure. Companies watch profit. Herders watch income. Each part may be locally rational while the system as a whole remains irrational. What is interesting about Talatan is that it places several accounts that used to be separate onto the same table.

Is maximizing electricity generation the only objective? No. If slightly higher construction costs reduce future spending on desert control, fire prevention, and vegetation clearing, the overall balance may be better. Is more vegetation always better? Again, no. Too much grass may obstruct equipment and increase fire risk. Do more sheep always mean more income? Not necessarily. Once grazing exceeds the carrying capacity of the land, today’s additional income becomes tomorrow’s ecological debt. Good governance does not maximize a single variable. It searches for a workable balance within a system.

V

This may be the most important addition that public governance makes to the traditional philosophy of the unity of knowledge and action. In an individual, the unity of knowledge and action often emphasizes courage. At the national level, what matters even more is feedback. When an individual makes a mistake, the cost may fall mainly on that person. When a government makes a large-scale policy mistake, the cost may be borne by millions of people and may persist for many years. A country therefore needs institutionalized humility even more than an individual does.

Such humility does not mean indecision or fear of acting. It simply acknowledges a basic fact: no matter how intelligent a decision-maker may be, no one can know every consequence before action begins. The true reliability of a modern state does not lie in never making mistakes. No such state exists. What matters is whether mistakes can be discovered, whether, once discovered, they can be spoken about, whether, once spoken about, they can be corrected, and whether successful corrections can be retained as institutional memory.

This is where Talatan matters. When photovoltaic development began, people were thinking primarily about energy. Then the land told them there was also ecology. As the ecology improved, the grass told them there was also maintenance. Once the sheep arrived, the herders told them there was also livelihood. Once herders entered the system, the grassland told them there was also carrying capacity. Every problem that is solved generates another problem. That does not mean the earlier decision was a failure. On the contrary, this is how a complex society normally works.

Good governance does not solve every problem once and for all. It raises society’s capacity to solve the next kind of problem while dealing with the present one.

One of the most interesting capacities China has developed over the past several decades may be precisely this ability to learn by doing. It has certainly not succeeded every time, and this idea should never be used to romanticize every policy as an “experiment.” Experiments need boundaries. They must be capable of being stopped. Failures must be recorded. People who bear the costs must have channels through which they can speak. Successes must also be independently tested rather than automatically declared “experience” simply because they fit a grand narrative. Otherwise, the word “experiment” can become a convenient cover for bad policy.

A mature national system of knowledge and action therefore requires one final condition, and it is not a romantic one: facts must be allowed to matter more than face. If Talatan were permitted to have only one official story—“we designed everything brilliantly from the beginning”—then the case would lose its most valuable meaning, because that is not what happened. No one originally designed the complete photovoltaic–grass–sheep–herder–ecological-management loop that exists today.

What deserves respect is something else: people did something, the world produced an unexpected result, and instead of simply treating that result as a nuisance to be eliminated, they kept observing, changed their methods, and eventually turned the unexpected into a new possibility. Grass was not a main character in the original plan. Neither were sheep. But both eventually became part of the system. That is innovation.

VI

When we talk about technological innovation today, we quickly think of semiconductors, artificial intelligence, quantum computing, and robotics. All of these matter. But a country’s deeper capacity for innovation also includes another ability that rarely becomes a headline: when confronted with a problem that has no ready-made answer, can it allow reality to participate in producing the answer?

That may be Talatan’s most important lesson. Knowledge is not a complete map stored in the mind before action begins. Action is not a button pressed only after knowledge is complete. The real world is more like walking through darkness. We have maps, experience, instruments, and a sense of direction, but once we start moving, we still encounter rivers, landslides, and forks in the road that were never marked on the map. An intelligent person does not throw the map away, nor does he walk into a wall while staring at it. He looks up at the road. That is the unity of knowledge and action at the level of the individual.

A nation is no different. Planning matters. Strategy matters. Expertise matters. But if a country loses the ability to correct itself in response to reality, even the grandest plan may eventually become a paper world. By contrast, if a country can bring together the engineer’s measurements, the herder’s experience, the company’s costs, the scientist’s research, local experiments, and the central government’s long-term strategy, then what it gains is more than a power station and more than a grassland. It gains something more valuable: the capacity to keep producing knowledge.

What China should value most about Talatan today is therefore not a “world number one” title, nor even the catchy name “solar sheep.” Talatan reminds us that a truly intelligent person is not simply someone who knows a great deal, but someone who allows action to keep correcting understanding. A truly strong country, likewise, is not defined only by the amount of capital, technology, or engineering capacity it possesses. It is also defined by whether it can turn countless forms of local experience, feedback from different levels, and a constantly changing reality into the knowledge needed for the next round of decisions.

Wang Yangming left the individual with a question: can knowledge and action truly become one? More than five hundred years later, Talatan has unintentionally turned that into another question: How can a modern country with a vast population, an immense territory, and a highly complex industrial system allow knowledge continually to enter action, while action continually produces new knowledge?

That question matters far more than the simple fact that “grass grew beneath the solar panels,” because what ultimately determines a country’s future is not how many answers it already possesses, but whether it can continue to learn when old answers stop working and new problems appear.

What is most worth replicating from Talatan, in the end, is not photovoltaic power, not sheep, and not any single technique of desert control. It is this capacity to learn. The deepest form of the national unity of knowledge and action is to make sure that reality is always allowed to speak.

Cognition expands your field of vision. Opportunity changes your position. Action brings feedback. Capability compounds.

A post circulating on social media titled “I Was Genuinely Shocked by Sam Altman’s Level of Thinking” does capture something worth examining. When Altman looks at technological change, he rarely stops at the question, “How useful is this tool today?” He is more likely to ask: if a cost curve keeps falling and a capability keeps improving, what will suddenly become possible five or ten years from now? That may be more revealing than asking how much Altman himself understands about neural networks.

Altman has spent much of his career in two unusual positions. In one, he watched large numbers of startups try to build themselves from nothing. In the other, he has been directly involved in changes in the capability and cost of artificial intelligence. Put together his writing, interviews, and entrepreneurial experience over the past decade or more, and a fairly clear line of thought emerges: cognition determines what you can see; opportunity determines where you place your bet; action determines whether an idea enters reality; capability determines how much you can ultimately amplify the opportunity. Altman has never formally presented these four words as a theory, but they help explain many of his otherwise scattered judgments.

One of Altman’s most important habits is that he does not treat the present as a smaller version of the future. When people encounter a new technology, their first instinct is often to list what is wrong with it today: AI hallucinates, AI-generated writing can sound artificial, its reasoning can fail, and professional work still requires human verification. From this they conclude that AI cannot replace professionals. None of those observations is necessarily wrong. The problem is that they answer the question, “What can AI do today?” and then use that answer to predict what AI will change several years from now. The missing variable is the speed of change. Altman is less interested in how intelligent a particular model looks at this moment than in what happens if model capability keeps rising while prices and barriers to use keep falling.

Technologies that transform industries usually do not move overnight from “useless” to “perfect.” They gradually cross a threshold at which they become cheap enough, convenient enough, and reliable enough. Photography had existed for more than a century before smartphones drove the marginal cost of taking a photograph close to zero and suddenly turned much of the world into photographers. Video conferencing existed for years before bandwidth, cameras, and software became cheap and good enough to make remote work part of ordinary life. Computers began as expensive machines owned by a small number of institutions. As prices fell, they moved from computing centers to desks and eventually into people’s pockets. Artificial intelligence may follow a similar path.

A person who uses AI for market research today still has to verify the results. That does not mean market research will remain unchanged. The more useful question is this: if the first round of research that once required five people and two weeks can eventually be done by one person in a day, what kinds of small companies suddenly become capable of entering overseas markets? If information gathering, competitor analysis, and multilingual research once affordable only to large consulting firms fall in cost by 90 percent, who gains a new opportunity? The key question is not whether AI still makes mistakes. It is whether the minimum economic scale of an activity has changed.

That leads to the second word: opportunity. Major opportunities usually do not come from discovering a secret that nobody else knows. More often, everyone sees the same technology, but only a few people seriously think through what happens if it keeps developing. In 2007, many people saw the smartphone. But seeing a smartphone was very different from seeing mobile payments, ride-hailing, short video, food-delivery platforms, mobile advertising, and the sharing economy. Since 2022, nearly everyone has seen generative AI. The real dividing line is no longer whether someone has heard of ChatGPT. It is whether they can push the logic one step further. If first drafts become cheaper, what becomes more expensive? If code becomes cheaper, where does the bottleneck in product development move? If translation becomes cheaper, how far does the barrier to cross-border business fall? If information search becomes cheaper, does judging what is true become more valuable? If everyone can generate content, what becomes scarce? The answer may not be more content. It may be credibility, judgment, proprietary data, customer relationships, brand, and real-world experience.

That is why the claim that “the real value is not the tool, but the capability the tool amplifies” gets at an important feature of the AI economy, even though it is not an Altman quotation. Give AI to someone who does not understand an industry, and that person may simply produce mediocre answers faster. Give the same AI to someone who understands the product, understands the customer, and is willing to test the results, and that person may be able to test hypotheses, organize feedback, analyze competitors, and adjust the product far more quickly. The tool is the same. The result is completely different.

Future competition in work may increasingly look like multiplication:

Human capability × AI leverage

If the underlying judgment is close to zero, multiplying it by ten still does not produce much. But if someone already possesses rare industry knowledge, valuable data, distribution channels, or trusted relationships, amplifying those capabilities several times can produce a completely different result.

The better question, then, is no longer simply, “Will AI replace my job?” It is: Which part of this job is becoming commoditized, and which part is becoming more valuable? A researcher may once have spent most of the time gathering information. If information becomes cheap, defining the right question becomes more important. Programmers once spent enormous amounts of time writing routine code. If code generation becomes cheap, architecture, product judgment, security, and understanding users may become more valuable. Marketing professionals once gained an advantage from being able to produce content. If everyone can generate content, the person who truly understands why customers pay may become more valuable. Technological progress often works this way. It does not raise everyone’s value evenly. It reprices different kinds of capability.

The third word is action. Altman has long emphasized a very simple entrepreneurial principle: many questions cannot be answered until something enters the real world. A great many people think “figuring things out” is a stage that must be completed before action begins. They want to research every competitor, analyze every possibility, write the complete plan, identify the optimal path, and only then start. But the real world is not a math problem with an answer that can be solved on paper. Customers change. Technology changes. Competitors change. Search rules change. Policy changes. A “perfect plan” that takes six months to prepare may already be obsolete on the day it launches.

A more effective approach is often to identify the most important assumption, build the smallest version that can test it, put it into the real world, let reality prove you wrong, and revise. This is not anti-intellectual, and it is not a defense of recklessness. It simply recognizes something important: reality contains more information than the model in your head.

An entrepreneur sitting in an office debating whether customers will like a product has only a hypothesis. Whether customers actually pay produces new information. A writer wondering whether readers will understand an essay is still making a judgment. Giving it to dozens of real readers produces evidence. Action, then, is not merely execution. Action is itself a way of acquiring knowledge. That is the real value of “act first, then revise through feedback.” High-level thinking does not mean sitting in a room until the future has been completely worked out. It means forming a better hypothesis than most people, then letting reality test it faster.

The final word is capability, and this may be the most misunderstood concept in the AI era. Many people are now learning “how to use AI.” They master prompting, workflows, and automation tools, and assume they have gained a durable advantage. The problem is that tool-level skills may have a very short half-life. Today it may take a carefully designed ten-step prompt. Next year, a model may accomplish the same thing from one sentence in ordinary language. Today a workflow may require elaborate integrations. The next generation of products may simply include the function by default. If someone’s competitive advantage is merely being more familiar with a particular AI button, a product update can erase that advantage overnight.

The capabilities worth accumulating over the long term are those AI can amplify but cannot easily create from nothing: deciding which problems are worth solving; understanding how an industry really works; knowing the difference between what customers say and what they actually pay for; recognizing reliable evidence; knowing when a model is making things up; building a reputation that makes other people willing to work with you; organizing talented people; sustaining long-term projects; and continuing to think independently when everyone else is doubtful. The most important future capability may no longer be doing all the work yourself, but rather deciding what work is worth doing and organizing people, AI, capital, and information to get it done. That is a higher-order capability.

There is, however, an important limit to calling all this “Altman’s art of enlarging a life.” Without that limit, the argument easily slides into success mythology. Altman’s path is not something an ordinary person can reproduce simply by mastering four principles. He entered Stanford at a young age, joined the Y Combinator network early, and later found himself at the center of an unusually concentrated network of capital, technology, and talent in Silicon Valley. Family background, education, social networks, historical timing, and the ability to absorb risk all shape what a person can realistically do. Someone with no savings cannot “take bold risks” in the same way as someone with a financial cushion. Someone supporting three children cannot simply hear “seize the opportunity” and quit a job. Success is never only a competition in cognition. Resources, institutions, and luck matter too.

What is worth learning from Altman, then, is not how to imitate his life. It is a way of seeing the relationship between oneself and the world as something dynamic. The world is not a fixed background, and human capability should not be treated as a fixed asset. When a new technology appears, four questions can guide the response. What is it changing? That is cognition. What is becoming cheaper because of it, and what is becoming more valuable because it is scarce? That is opportunity. Can a real experiment be run at low cost? That is action. After that experiment, what should be accumulated that others will find increasingly difficult to replace? That is capability.

Once those four questions make a full circle, they return to the beginning. New capabilities reveal new opportunities. New opportunities force people to learn new things. New actions generate new information. New information changes the original cognition. That is where compounding begins.

Perhaps the most interesting thing about Altman is not whether his predictions about the future turn out to be right. He will certainly be wrong at times, perhaps very wrong. Anyone standing at the center of a technological wave has an incentive to imagine that the thing he is working on matters more than it eventually will. What is more interesting is the way he often combines two very different time horizons: he stretches the horizon when looking at trends, but shortens the cycle when taking action. Look at the future over five or ten years. Run the experiment today.

Most people do not need to predict the next technological revolution, and few need to build the next OpenAI. But they can train a more practical habit. When something new appears, do not immediately ask whether it is perfect today, and do not immediately ask whether it will replace you. Ask what it is making cheaper. Ask whom it is making stronger. Ask where you will stand if the trend continues for another three years. Then do one small thing and see whether your judgment was wrong.

Cognition is not about appearing intelligent. It is about seeing changes that others have not yet fully seen. Cognition has to become opportunity. Opportunity without action remains imagination. Action that does not build capability is merely busyness. What gradually enlarges a life may be this cycle:

Cognition expands your field of vision. Opportunity changes your position. Action brings feedback. Capability compounds.

AI is only the most powerful new variable in the current round of change. In the end, what determines how far a person can go is still not what tools they possess, but whether they can understand the world more accurately over time—and whether they can take the first step while the world is still uncertain.

Li Zide

Subscribe now

On July 24, 2026, China’s Ministry of Finance and State Taxation Administration issued Announcement No. 21, setting out a systematic framework for the individual income-tax treatment of offshore trusts（离岸信托）. The announcement may look highly technical. It deals with settlors, trustees, offshore entities, gains from property transfers, tax residency, and effective control. Yet its real impact may reach much further, touching the entire system of wealth arrangements that has taken shape among China’s high-net-worth individuals over the past two or three decades.

Pan Shiyi（潘石屹）stands almost exactly at the point where these changes meet.

The story of his wealth, and that of his wife Zhang Xin（张欣）, contains nearly every defining feature of China’s three decades after reform and opening: real estate, urbanization, Hong Kong listings, offshore companies, global investment, international education, overseas assets, and family wealth management. They came of age in an era when Chinese asset prices were rising rapidly and cross-border capital allocation was steadily expanding. In that sense, the “Pan Shiyi dividend”（潘石屹红利）was never simply about how much money Pan himself made. It was the dividend of an entire era.

Now some of the institutional conditions that made that dividend possible are changing.

The most important feature of Announcement No. 21 is that China’s tax administration is moving more explicitly beyond the formal boundaries of companies, trusts, and passports in order to identify who actually controls assets and who ultimately enjoys their economic benefits. When a resident individual transfers property into an offshore trust, the taxable gain is, in principle, calculated on the basis of the property’s market value at the time of transfer, minus its original cost and reasonable expenses. Income subsequently generated by the offshore trust, or by offshore entities it controls or manages, may also have to be declared and taxed annually even if that income has never actually been distributed to the individual.

This reflects an important change in the way tax liability is understood. In the past, the key question was often: Has the money actually been paid into the individual’s own account? Increasingly, the question is becoming: Whose money is it in substance? Who put the assets there, who can decide how they are invested or disposed of, and who ultimately receives the economic benefit?

If the answer still points back to a Chinese tax resident, then in some circumstances the tax obligation may not disappear simply because a Cayman Islands（开曼群岛）company, an offshore trust, a nominee, or some other legal entity stands in between. This is what is meant by “look-through” taxation（穿透）.

Announcement No. 21 goes further, touching on another issue long central to cross-border wealth planning: foreign nationality. Acquiring foreign citizenship or long-term or permanent residence abroad does not automatically sever an individual’s relationship with China’s personal income-tax system. If the person’s principal economic interests remain in China, that person may still, under the relevant conditions, be treated as a resident individual with domicile in China.

So a passport will increasingly be only one factor in determining a person’s tax position. Other questions become more important: Where does the person actually live? Where is the family based? Where are the businesses? Where does the main wealth come from? And who actually controls the assets?

In a globalized capital system, a person can hold one passport, register a company in another country, establish a trust in a third jurisdiction, and own investments across several continents. Tax authorities, however, are increasingly trying to piece these scattered legal relationships back together and reconnect them to a single person. That is what makes Announcement No. 21 important. It does more than impose another tax rule. It begins to redefine what counts as “your money.”

I

Pan Shiyi is therefore an unusually revealing figure through which to understand this moment.

The rise of SOHO China（SOHO中国）was itself part of China’s urbanization and real-estate boom. Beijing（北京）and Shanghai（上海）expanded rapidly, land prices rose, and office buildings became an important class of asset. Pan Shiyi and Zhang Xin seized that moment and developed a series of highly recognizable commercial properties. Later, they caught another wave: the globalization of capital. Chinese companies listed in Hong Kong（香港）, offshore holding companies became common, and international banks, funds, and trust institutions entered the wealth-management world of Chinese entrepreneurs.

For the first time, large numbers of Chinese businesspeople were operating in a world where a company could make money in Beijing, list in Hong Kong, build a corporate structure in the Cayman Islands, buy property in New York（纽约）, donate money in Boston（波士顿）, and manage assets in London（伦敦）. Today this may sound ordinary. Two or three decades ago, it represented a profound institutional change. Chinese private wealth had, for the first time, truly crossed national borders on a large scale.

This is the first key to understanding Pan Shiyi.

He was not the traditional kind of rich man who locked his money in a safe at home. The age he lived in taught entrepreneurs that assets could move, legal entities could be layered, ownership could be divided, and wealth could be allocated globally. There were boundaries between the individual and the company, between China and the outside world, between ownership and beneficial interest, and among settlors, trustees, and beneficiaries.

Those boundaries are part of modern commercial law. But boundaries can protect legitimate transactions, and they can also create room for tax planning. Over the past several decades, tax authorities around the world have faced the same basic problem: if one person’s wealth is spread across a dozen companies, several trusts, and multiple countries, can the traditional tax system still find that person?

The answer is changing.

II

The changes in the wealth of Pan Shiyi and Zhang Xin over the past two decades have therefore acquired a particular symbolic meaning.

Publicly available information shows that SOHO China continued to sell some of its assets after the peak of the property boom. At the same time, Zhang Xin participated in commercial real-estate investments in the United States, including prominent properties in New York. Both facts can be established. But an essential evidentiary boundary must be maintained: the fact that domestic asset sales and overseas investments occurred during overlapping periods does not prove that each sum raised in China flowed directly into a particular overseas account. Nor can a complete path of capital movement be drawn merely from chronological coincidence.

Commercial judgment and factual judgment must remain separate.

The frequently cited figure of 12.8 billion yuan also requires caution. Some financial media have estimated, on the basis of SOHO China’s past dividend payments and the couple’s shareholding, that the shares controlled by Pan Shiyi and Zhang Xin generated cumulative cash dividends of roughly 12.8 billion yuan. That figure has a basis for discussion. But “having received roughly 12.8 billion yuan in dividends over time” is entirely different from saying that “a particular offshore trust now contains 12.8 billion yuan.”

After money is received, it may be invested or spent. It may still be held, or it may have been converted into other assets. Taxes may already have been paid on some of it, while other portions may involve different tax years, different tax-residency statuses, and different legal entities. Without complete trust documents, records of asset changes, tax-residency information, and historical tax filings, outsiders cannot accurately calculate how much tax Pan Shiyi and Zhang Xin might owe.

That distinction matters. Tax law cannot be based on guesswork, and wealth cannot be treated as evidence of wrongdoing simply because it is large.

If tax authorities later determine, on the basis of law and fact, that taxable income exists, the tax should be collected. If tax evasion is established, penalties should follow. But before the facts are known, no one should issue a hypothetical tax bill of tens of billions of yuan simply because the person involved is wealthy. Taxation is not a matter of emotion.

III

Yet even if we do not know how much tax Pan Shiyi may ultimately owe, he remains an excellent case through which to understand Announcement No. 21, because the “Pan Shiyi dividend” depended precisely on boundaries.

In the past, registering a company offshore could make financing and listing easier. Establishing a trust could serve wealth succession, asset protection, and family governance. Investing through an overseas company could diversify geographic risk. Moving one’s residence could reorganize family life and wealth planning. None of these acts is inherently unlawful, and an offshore trust is not, by itself, a tool for tax evasion. Trusts are widely used around the world for ordinary family wealth management, charitable foundations, inheritance planning, and cross-border investment.

The problem appears elsewhere.

When the gap between legal form and economic reality grows too wide, tax authorities begin to look beyond the name written at the top of a contract. They ask further questions: Who really put up the money? Who really makes the decisions? Who ultimately receives the money? Who bears the risks and enjoys the returns?

This has become one of the clearest trends in global taxation over the past decade or more. Company names can change. Trustees can change. Passports can change. But effective control and ultimate economic benefit are becoming harder to hide behind those formal structures.

Announcement No. 21 is an important extension of that trend into the taxation of private wealth in China.

IV

The “Pan Shiyi trap”（潘石屹陷阱）, then, does not mean that Pan’s earlier arrangements were necessarily unlawful. The real trap is that an arrangement that worked successfully under one institutional environment may face an entirely different regulatory logic twenty years later.

That is the heart of the issue.

One of the easiest mistakes for entrepreneurs to make is to treat temporary institutional conditions as permanent rules. Property prices will always rise. Cross-border capital movement will always become freer. Offshore structures will always retain the same tax effects. Once citizenship or residency changes, previous tax connections will automatically disappear. History repeatedly shows that none of these assumptions can be taken for granted.

Institutions change, and so does the capacity of the state.

Twenty years ago, a tax authority might not even have known which overseas companies a resident controlled. Today, global financial-account information exchange, bank compliance, beneficial-ownership registration, anti-money-laundering rules, and digital tax administration have transformed the information environment. In the past, information was scarce. Now what is becoming scarce is the ability to remain invisible.

That is the fundamental change confronting Pan Shiyi’s generation of wealthy Chinese. They accumulated their fortunes during an era in which globalization kept opening borders. Now, as they enter the stage of intergenerational wealth transfer, they find tax authorities around the world reconnecting those borders.

One era opened the doors. Another has begun asking who lives behind them.

V

Announcement No. 21 also contains a highly practical provision dealing with existing arrangements.

Under the announcement, certain offshore-trust tax matters from earlier periods are given a ninety-day filing window. Where the relevant conditions are met, taxpayers who come forward within that period may avoid the corresponding late-payment surcharge. The message to high-net-worth individuals is clear: some offshore wealth structures that have long existed in a gray area now need to be reviewed.

For tax lawyers, accountants, family offices, and private banks, the next step is unlikely to be a matter of filling out a single form. They may need to review an entire wealth structure: Who is the tax resident? Who has effective control? What kind of trust is involved? When were the assets transferred into it? What was their original cost? How much income has been generated over the years? How much tax has already been paid overseas? Has the status of any family member changed? Should the existing structure be kept at all?

These questions will directly shape the next round of wealth allocation among China’s high-net-worth families. That may prove to be the most important economic consequence of Announcement No. 21.

How much additional tax the authorities collect this year—tens of billions of yuan or hundreds of billions—certainly matters. But something else may matter more: a group of people with very large fortunes will now reconsider where they want to keep their assets over the next ten years.

Tax policy does not merely collect yesterday’s money. It also changes where tomorrow’s money goes.

VI

This brings us to a question much larger than Pan Shiyi himself.

The state has good reason to close tax loopholes. The monthly income of an ordinary salaried worker is relatively easy for the tax system to see. If someone worth tens of billions of yuan can place a few more companies and trusts between himself and his assets, adopt several legal identities, and thereby leave large amounts of income outside the tax system for long periods, the tax system will eventually lose its sense of fairness.

No society can sustain a system in which those who are easiest to see are the easiest to tax, while those wealthy enough to hire sophisticated lawyers are the easiest to make disappear. If the ability to escape taxation rises with wealth, an increasing share of the public burden will fall on wage earners and middle-class families who cannot move their assets across borders.

That is one of the most basic questions of tax fairness.

The wealthy have every right to reduce taxes legally, arrange their wealth, and invest internationally. But access to more expensive lawyers and accountants should not amount to the purchase of a kind of exemption from tax law that ordinary people could never afford.

That point is not especially controversial. The harder question comes next.

VII

The state has another responsibility besides collecting tax: it must keep the rules predictable.

This is often lost in emotionally charged debates over taxation.

Capital is practical. It looks not only at tax rates but also at whether the rules are likely to change abruptly. If a high-net-worth family expects its tax burden to rise, it may spend less or reallocate its investments. If it expects a particular tax-avoidance strategy to be closed, it may redesign its structure. These are normal economic responses.

But if people cannot tell whether a wealth arrangement that is lawful today will be treated completely differently ten years from now, they are no longer thinking only about tax rates. They are thinking about institutional risk. Their responses may then become much larger: entrepreneurs may reduce long-term domestic holdings, change tax residency earlier, move the next generation abroad sooner, or place the intellectual property, holding structure, and even the founding location of a new company outside China from the beginning.

This creates a genuine policy dilemma.

Governments naturally want more tax revenue. They also want to close loopholes, keep capital at home, encourage entrepreneurs to invest, and, if possible, prevent wealth from flowing abroad. The difficulty is that these goals cannot always be maximized at the same time.

Policy always has a cost.

There is no system in the world that can close every loophole, collect every possible tax dollar, and still leave capital completely unchanged in its behavior. That is what makes tax policy difficult.

VIII

For that reason, three things must be kept separate when discussing Announcement No. 21: legal tax planning, illegal tax evasion, and new tax obligations created by changes in policy. They are not the same thing.

A trust is not a criminal instrument. Overseas assets are not evidence of wrongdoing. Nor does earning money in China mean that a person’s wealth must remain in China forever. Cross-border investment, international asset allocation, and family migration are all normal features of a modern economy.

The idea that “money earned in one place must stay there forever” may express a simple moral intuition, but it is not a complete principle of modern tax law. Tax liability is determined by tax residency, the source and nature of income, the nature of the assets, control relationships, and the law itself.

The same applies to Pan Shiyi.

Every yuan of tax legally owed should be paid. But tax that is not legally owed should not be imposed simply because public opinion dislikes a wealthy person. That is one of the most basic boundaries of the rule of law.

IX

The “Pan Shiyi trap,” then, has two levels.

The first belongs to entrepreneurs. The greatest mistake is to believe that a legal structure found today can permanently protect wealth against tomorrow’s institutional changes. Wealth planning is never a one-time exercise. Changes in tax law, international rules, the status of family members, and the nature of assets can all make yesterday’s effective structure obsolete.

The second belongs to government. Governments, too, can fall into the illusion that tighter regulation will automatically deliver more tax revenue, greater fairness, more capital, and stronger confidence all at once.

It may not.

The stronger a state’s regulatory capacity becomes, the more important stable rules become. The more capable the state is of looking through the shells of companies, trusts, and passports, the more clearly taxpayers need to know when that power will be used, how far it will reach, and on what legal basis.

Otherwise, regulatory power itself becomes a kind of risk premium.

What capital fears most is not necessarily high taxation. Many countries have high tax burdens and still attract long-term capital. What capital fears more is not knowing how the rules will be applied tomorrow.

Institutional certainty is itself an economic asset.

X

Pan Shiyi’s story therefore carries a particular meaning for this moment in Chinese history.

When he was young, he entered a China whose borders were steadily opening. Cities expanded. Real estate became a machine for creating wealth. Private companies grew quickly. Chinese firms entered Hong Kong’s capital markets. Capital went global. For the first time, Chinese families began allocating wealth on a large scale in New York, London, Hong Kong, and Singapore（新加坡）.

One door after another opened. Pan Shiyi understood those doors early, and that is how he captured the “Pan Shiyi dividend.”

Thirty years later, China is redefining the rules behind those doors. Behind a company, regulators now look for the controller. Behind a trust, they look for the beneficiary. Behind a passport, they look for tax residency. Behind an overseas asset, they look for the person who ultimately holds the economic interest.

This is not unique to China. Tax systems around the world have been moving from legal form toward economic substance, from individual accounts toward effective control, and from nationally segmented information toward cross-border information exchange.

But in China, this shift is taking place at an unusually sensitive moment. The property boom is over. Local governments face fiscal pressure. Household wealth is growing more slowly. Business confidence still needs rebuilding. Cross-border capital flows are under closer scrutiny. At the same time, China must deal with new questions created by concentrated wealth, tax fairness, and the globalization of private assets.

The ultimate test of Announcement No. 21, then, is not simply whether China can find money held in offshore trusts. The real test is whether China can establish a new balance: one in which wealth finds it hard to escape the tax system without feeling compelled to leave China; one in which the rich bear their proper share of public obligations without wealth itself being treated as a moral offense; and one in which the state has the power to look through complex ownership structures while entrepreneurs can still predict the basic rules ten years ahead.

That is where the real difficulty lies.

The “Pan Shiyi dividend” belonged to an entrepreneur who learned early how to use the borders opened by the age of globalization.

The “Pan Shiyi trap” appears more than twenty years later, when his generation may suddenly discover that the borders are still there, but the state has learned how to look across them and find the person behind them.

And the real question for China’s economy is not how much money can ultimately be collected from wealthy people like Pan Shiyi. It is a much larger question:

Once the state has gained the ability to find wealth, how can it make sure that wealth still wants to be created there?

That is the next question China must answer as it moves from the “Pan Shiyi dividend” to the “Pan Shiyi trap.”

Subscribe now

Li Zhide

On May 24, 2000, after a fierce debate, the U.S. House of Representatives voted 237 to 197 to approve legislation granting China permanent normal trade relations. On October 10, President Bill Clinton signed the U.S.-China Relations Act into law as Public Law 106-286. China had in fact enjoyed normal trade relations with the United States since 1980, but under the framework of the Jackson-Vanik Amendment that status had been subject to periodic review in the American political process. The 2000 law changed that. Once China joined the World Trade Organization, the United States could grant it permanent, nondiscriminatory normal trade treatment without reopening the issue year after year. After China joined the WTO in December 2001, this arrangement became part of the basic legal structure of U.S.-China economic relations.

The logic at the time was not difficult to understand. Mainstream American policymakers believed that integrating China more deeply into the global trading system would produce at least three results: American companies would gain access to a vast market, American consumers would receive more inexpensive goods, and China itself would alter its economic behavior as it entered a rules-based global trading order. The Clinton administration regarded permanent normal trade relations, or PNTR, as an important part of its broader strategy of engagement with China. Looking back on that history in 2004, U.S. Trade Representative Robert Zoellick still emphasized that PNTR had emerged after years of annual congressional debate and ultimately became an institutional arrangement supported by a bipartisan majority.

A generation later, the political landscape has changed almost beyond recognition. On August 21, 2026, the U.S. International Trade Commission is expected to release Investigation No. 332-609, bluntly titled “Effects on the U.S. Economy of Revoking China’s Permanent Normal Trade Relations Status.” The commission is no longer being asked whether another 10 or 25 percent tariff should be imposed on selected Chinese products. It is examining a more fundamental question: what would happen to U.S. trade, industry, prices, and sources of supply if the United States revoked China’s PNTR status? The investigation also includes an alternative scenario examining what would happen if Congress ended PNTR and phased in higher tariffs over five years on certain products connected to national security. Because the schedule is compressed, the commission is not holding a public hearing and is relying instead on written submissions.

The report will not decide the issue for Congress. A Section 332 investigation is a fact-finding exercise. It provides independent analysis and does not make policy recommendations. There is therefore no factual basis for saying in advance that the August 21 report will “certainly endorse revoking China’s PNTR.” What matters is that American politics has turned a question that would have seemed almost unimaginable more than twenty years ago into a formal government research project: Should the legal foundation laid in 2000 for U.S.-China trade still be preserved? What deserves attention is not merely how much higher Washington may want tariffs to go. The deeper shift is that restrictions on trade with China are moving from tools that presidents can adjust toward institutional arrangements that some members of Congress are trying to lock into law.

Looking back now, the point is not to prove that the people of 2000 were “foolish” or that those of 2026 are “right.” History is far more complicated than that kind of retrospective judgment. After China entered the WTO, U.S.-China trade expanded rapidly. American consumers, retailers, multinational corporations, and many export industries benefited. At the same time, manufacturing employment, the bilateral trade deficit, industrial relocation, intellectual-property disputes, subsidies, and dependence on Chinese supply chains became increasingly central to American political debate. By Donald Trump’s first term, the old assumption that the United States could bring China into the system and then encourage it to change within that system was steadily giving way to another conclusion: the United States needed to change the terms under which China participated in the system.

The Section 301 tariffs imposed beginning in 2018 marked a turning point. But one distinction is often overlooked. The repeated imposition of Section 301 tariffs on Chinese goods did not mean that China had already lost PNTR. The two systems continued to coexist. Chinese products still entered under the normal trade relations tariff schedule as their base rate, while some products were subject to additional Section 301 and other duties. As of 2026, the Congressional Research Service still notes that the tariffs imposed on Chinese goods under Section 301 of the Trade Act of 1974 during the first Trump administration remain in force.

In other words, much of the trade war of the past eight years has involved adding more and more locks to the same house. What some members of Congress are now considering is whether to touch the foundation itself.

The distinction is visible in the U.S. Harmonized Tariff Schedule. Column 1 applies to the overwhelming majority of countries that enjoy normal trade relations with the United States. Column 2 applies to the small group that do not. The countries currently listed in Column 2 are Belarus, Cuba, North Korea, and Russia. Moving China from Column 1 to Column 2 would therefore carry political meaning far beyond an ordinary tariff increase. It would mean that the United States no longer placed China in the same legal category as most of its normal trading partners.

And Congress already has legislative language on the table. On January 23, 2025, Representative John Moolenaar and Democratic Representative Tom Suozzi introduced H.R. 694, the Restoring Trade Fairness Act. Senator Tom Cotton and others introduced the Senate version, S. 206. Both bills center on ending China’s PNTR status. The House bill does not simply say that China would thereafter fall under Column 2. It goes further by creating a tariff structure specifically for China: beginning from Column 2 rates, it would establish a minimum ad valorem tariff of 35 percent where the existing rate is lower, and for certain products listed in Section 10, a minimum rate of 100 percent where the existing rate falls below that level. Most of the increases would be phased in over five years. The bill also changes the rules governing the $800 de minimis exemption and restricts access to that treatment for goods from designated countries.

But “a bill has been introduced” is not the same as “the United States has decided to revoke PNTR.” H.R. 694 and S. 206 are still proposed legislation, not law. The House version has bipartisan sponsorship. The Senate bill, by contrast, was introduced by Republicans including Cotton, Jim Banks, and Josh Hawley. Congress has not yet reached a bipartisan consensus on revoking China’s PNTR status.

That is why the real question is not whether these bills will pass tomorrow. The more important question is why the idea of revoking PNTR has moved from the political margins into formal economic assessment and actual legislative text.

The Supreme Court added another institutional layer to the debate on February 20. In Learning Resources v. Trump, the Court ruled that the International Emergency Economic Powers Act, or IEEPA, did not authorize the president to impose tariffs. The decision directly limited one legal path through which a president might use a declared “national emergency” to levy sweeping duties.

But that does not mean that presidents can no longer impose tariffs, nor does it mean that “congressional legislation is the only route left.” Presidents still possess authority under other trade laws previously enacted by Congress. Section 301 addresses unfair foreign trade practices. Section 232 concerns national security. Section 201 deals with import surges. Each provides a different degree of executive authority over tariffs. In fact, in 2026 the Office of the U.S. Trade Representative is still using Section 301 investigations to support new trade actions, while the existing China tariffs are undergoing another four-year review.

The more accurate conclusion is this: the Supreme Court ruling again exposed the legal limits of tariff measures based on executive power, while a congressional decision to revoke PNTR would have an entirely different degree of institutional durability.

Executive actions taken under delegated authority can change because of court rulings, policy revisions, exemptions, or a new administration. Laws passed by Congress can also be amended, of course, but reversing them normally requires another legislative process. The political transaction costs are entirely different. This is what makes the possibility of “writing the trade war into law” so consequential.

A tariff is, at one level, a price. A 25 percent tariff imposed today can become 10 percent tomorrow, or disappear after the two sides reach an agreement. PNTR, however, is first of all a matter of legal status. It answers a different question. Not “How much duty does this Chinese product pay today?” but rather: “What category of country is China in under American trade law?”

Once that status changes, businesses no longer face only short-term tariff risk. If an American company believes a tariff may disappear in two years, it may wait. If it believes a higher tariff has been written into law by Congress and normal treatment can return only through another legislative act, it will reconsider suppliers, factories, inventories, sourcing locations, and long-term investment. Vietnam, Mexico, India, or production within the United States may attract more capital as a result. Once such supply-chain shifts occur, they create new economic constituencies. Those constituencies, in turn, may support keeping the policy in place.

This is institutional path dependence:

law changes expectations → expectations change investment → investment changes supply chains → supply chains create new political interests → those interests reinforce the original law.

The hardest thing to reverse, therefore, may not be a particular tariff rate. It may be the industrial and political structure that forms around high tariffs.

None of this would be costless. Revoking China’s PNTR status would not magically eliminate American demand for Chinese goods. Some imports could shift to third countries. Some production could return to the United States. But some products could also become more expensive because of China’s cost advantages, scale, and dense supplier networks. Higher import costs for businesses, higher consumer prices, the cost of Chinese intermediate goods used by American manufacturers, and possible Chinese retaliation against U.S. agriculture and other export sectors would all enter the final accounting. That is precisely why Investigation 332-609 asks the U.S. International Trade Commission to examine trade, domestic production, prices, and sources of supply, rather than merely calculating how many American firms might be protected by higher tariffs.

This is what gives August 21, 2026, its real importance. It will not be a starting gun. Nor will it prove that revoking PNTR has “locked and loaded.” What matters is that the U.S. government is, for the first time in such a formal way, calculating the economic consequences of a post-PNTR era with China.

Twenty-six years ago, Washington was doing a different calculation: what might the United States gain by giving China permanent normal trade treatment and bringing it more fully into the global market?

Twenty-six years later, it is beginning to calculate another: what would the United States pay, and what might it gain, if that treatment were withdrawn?

The change in the question may matter more than the eventual answer.

Washington in 2000 was largely asking:

How can China be integrated more fully into an international trading system led by the United States?

In 2026, a growing number of people in Washington are asking:

How can the United States reduce its economic dependence on China while raising the cost of Chinese access to the American market?

That is the real historical turn of the past twenty-six years.

Whether Congress ultimately revokes China’s PNTR status remains uncertain. American businesses, consumers, importers, retailers, agricultural states, and those worried about inflation and supply-chain costs will all take part in the political struggle. The USITC report may itself identify substantial domestic costs for the United States. But something else has already happened. The question of whether China should continue to enjoy permanent normal trade relations has once again become something American politicians can openly debate, government agencies can formally calculate, and members of Congress can directly legislate.

That alone marks a threshold.

Since 2018, the central logic of the U.S.-China trade war has still been to raise the price of Chinese goods entering the United States while remaining within the broader legal structure established earlier. The debate over revoking PNTR goes one step further. It begins to touch the legal status of the bilateral trade relationship itself.

Tariffs can still be negotiated. Laws can also be changed, but not as easily as an executive order.

The real question to watch, therefore, is not whether the August 21 report contains a dramatic average tariff figure or a particularly hawkish sentence. It is whether American politics is forming a new long-term consensus: that economic competition with China should not depend on one president, and should not disappear when that president leaves office, but should instead be embedded in U.S. trade law, industrial policy, and the national-security system.

Only if that process is completed will the nature of the U.S.-China trade war truly change. It would no longer be primarily a negotiation over how high tariffs should be. It would increasingly become an institutional contest over how much economic distance the two countries should maintain.

In 2000, the United States laid an important piece of domestic legal foundation for China’s entry into the global trading system.

In 2026, Washington is seriously considering whether to tear it up—and return to the world that existed before 2000.

Li Chengpeng (李承鹏)

The door has truly slammed shut. I just heard that China now has something people are calling an “exit tax.”

I looked into it. An exit tax is a special tax regime imposed on people who give up their citizenship or long-term residency. The United States, Canada, Germany, France, Australia, and Britain all have comparable systems. But they come with strict legal procedures and clear standards for taxation. In some cases, the tax applies only to corporate shares rather than an individual’s entire assets.

On paper, China does not have an “exit tax.” But in practice, the revised Individual Income Tax Law that took effect in 2019 added Article 13: “Where a taxpayer cancels his or her Chinese household registration because of emigration abroad, the taxpayer shall settle all taxes before the cancellation of household registration.” This means that before leaving, the taxpayer must file with the tax authorities, pay all outstanding taxes, and, where applicable, pay late fees and penalties.

And now the net is closing in so tightly that it is becoming very difficult to escape. In July 2026, the Ministry of Finance issued new rules on the individual income taxation of offshore trusts. The rules make clear that when a tax resident becomes a nonresident, a mandatory tax settlement is required. Public-security and tax authorities are also being brought into a stricter monitoring system, with particular attention paid to people who “emigrate without moving their assets.” The aim is full look-through tax enforcement.

This is the predicament now facing China’s middle class and everyone above it. When people first made their money, these rules did not exist. Then Article 13 was added to the Individual Income Tax Law in 2019. Seven years later, in 2026, the machinery is finally being put fully into operation. For seven years, people went on dreaming about “poetry and distant places,” telling themselves: Surely it won’t come to that. Surely the state won’t do anything to me.

How naïve. So what is your fate now?

———————————————————————

Here is the latest development. The combination of measures keeps coming. The Chinese government has begun levying individual income tax on mainland residents who purchase insurance policies in Hong Kong. Tax authorities in Beijing and Hangzhou have already imposed a 20 percent individual income tax on policy dividends and interest earned on prepaid premiums.

As soon as the news broke, shares of the internationally known insurer Prudential plunged as much as 13 percent during trading, while HSBC fell 6 percent. People in the industry have questioned the legal basis for the move. China’s current individual income tax law does not explicitly list “insurance-policy dividends” as a taxable category, they argue, so treating them as ordinary “dividends and profit distributions” may lack a clear basis in higher-level law.

But these objections are weak. As the Chinese government increasingly closes the country off while pursuing taxes on overseas income, Chinese investors are finding that there may be no way to walk away untouched. The next step will likely be a more aggressive retrospective pursuit of taxes on income from domestic assets.

The gate is being shut so the dogs can be beaten inside it. Those who spent the past few years ignoring public affairs, concentrating only on making money and dreaming of “poetry and distant places,” should wake up.

————————————————————

The Chinese state’s control over the free movement of ordinary people has a very long history. More than 2,300 years ago, Shang Yang (商鞅) wrote in The Book of Lord Shang (《商君书》): “If the people are not allowed to move about freely, they can be made ignorant.”

In plain language: keep ordinary people from moving around, and they will become easier to keep ignorant. He also argued that knowledgeable people should not be allowed to travel freely, because if they moved around and mixed with others, ignorant peasants might become wiser, making the state harder to govern.

So restrictions on movement were never simply about controlling people’s legs. They were about controlling their minds. Once people begin to move, they begin to learn. And once they become smarter, how is the ruler supposed to control them?

Shang Yang is often associated with five techniques for ruling the people: keeping them ignorant, weak, poor, exhausted, and uniform. In fact, a sixth should be added: keeping them locked in place. Shang Yang had six techniques of control.

——————————————————

Ordinary Chinese people were allowed to leave the country only after the Anglo-French forces fought their way into China. Article V of the 1860 Convention of Peking between China and Britain (《中英北京条约》) explicitly established the right of ordinary Chinese people to travel abroad. In plain language, it meant this:

If Chinese people wished to go abroad, whether to Britain or another country, whether alone or with their whole families, they were free to do so. Once they boarded a British ship, they were not to be stopped. The original Chinese text used four characters: 毫无禁阻—“without any obstruction whatsoever.”

So there it is. The first time the legality of Chinese people traveling abroad was written into a legal document, it was because the British forced the Xianfeng Emperor (咸丰皇帝) to put it in writing at gunpoint.

————————————————————

Chinese people have long lived with the illusion that the country’s leaders want to make the economy prosper because they want to please the people. But that assumption does not necessarily hold.

There is a political scientist named Bruce Bueno de Mesquita (布鲁斯·布恩诺·德·梅斯基塔) who saw through this problem with unusual clarity. In The Dictator’s Handbook (《独裁者手册》), he developed a cold-blooded framework known as selectorate theory. A ruler survives not by pleasing the population as a whole, but by keeping the smallest possible “winning coalition” satisfied. In other words, do not spend too much money improving the lives of ordinary people. Under an authoritarian system, citizens who are well fed, comfortably housed, and no longer dependent on what the ruler gives them can become dangerous.

When people are starving and close to revolt, then by all means develop the economy a little. Let Liu Yonghao (刘永好) raise pigs, let Liu Chuanzhi (柳传志) make computers, let Jack Ma (马云) build Taobao. But once you become big, strong, and successful enough to pose a threat to political control, then you have to be taken down.

That is what this latest closing of the country is about. Lock down the money. Lock down the technology. Lock down the people. A perfect closed loop. Your factory belongs to the state. Your company belongs to the state. Your money belongs to the state. You belong to the state. And your next generation belongs to the state as well.

——————————————

Chinese people have never truly enjoyed freedom of movement. From the travel certificate known as a yan (验) that Shang Yang’s system required for lodging, to the luyin (路引), or travel permits, of the Ming and Qing dynasties, to the work-unit introduction letters required after 1958, and on to today’s border-control watch lists, the instruments used to control our movement have changed over 2,300 years. Bamboo slips became pieces of paper. Paper became computer databases. Databases became AI facial recognition. But the underlying logic has not changed for a single day:

If you want to leave, you need my permission.

A PARTY that claims to have overthrown the old imperial order is still using the methods of emperors.

The principle could hardly be clearer. Yet after the new exit-and-entry regulations were announced, there were still plenty of idiotic takes. China’s public sphere has always resembled two skyscrapers rising into the clouds. Call one Building A and the other Building B. Those who think the new exit-and-entry rules violate basic human dignity live in Building A. Those who take pleasure in seeing the new rules punish wealthy people live in Building B. Let us hear what people in Building B have to say:

“I’m just an ordinary person. These rules are aimed at rich people with passports. Let them control them, arrest them, whatever. What does it have to do with me?” “If you haven’t done anything wrong, why are you afraid? Only people with guilty consciences are scared.” “I don’t have a passport and I don’t want to go abroad. China is so good—why leave? They should arrest everyone who wants to go abroad.”

Building B really is a remarkable structure. Then again, perhaps it is not so surprising. Some 900 million people in China have never been on an airplane, and 1.24 billion do not have passports.

From the 1950s through the end of the 1970s, nearly a million residents of mainland China risked drowning at sea, being eaten by sharks, or being shot by border guards as they desperately tried to escape to Hong Kong. The episode became known as the “Great Escape to Hong Kong” (大逃港).

In April 1978, Xi Zhongxun (习仲勋), on his way to take up a post in Guangdong, witnessed this mass flight firsthand. He said: “The problem is that our own living conditions are poor and we have been unable to solve it. How can we call these people illegal border crossers? These people are leaving the area, not fleeing the country. This is a contradiction among the people, not a contradiction between ourselves and the enemy. We cannot treat them as enemies.” He then gave an order: “Let them all go. We cannot simply arrest people. We must build up the mainland and make it a place they want to come back to.”

There was considerable humanity in Xi Zhongxun’s words. Yet forty-eight years later, under his son Xi Jinping (习近平), Li Qiang (李强) has promulgated the State Council Regulations on Exit and Entry Administration (《国务院关于出境入境管理的规定》), tightening controls just as another wave of people seeks to leave the country.

But there is a question. Back then, Chinese people fled because China was poor. Today China has the world’s second-largest GDP. So why are people trying to leave again?

————————————————

The State Council regulations state: “Persons who are not permitted to leave the country shall be promptly informed of the reason for the prohibition. In circumstances involving national security, criminal cases, and other such matters, the reason may be withheld.”

In theory, you should be told. In practice, you may not be.

Imagine that one day you are about to travel abroad. You slide your passport through the immigration counter. The young man behind the glass looks up at you and says coldly, “Go wait over there for a moment.”

Your heart suddenly starts pounding. You ask, “Why? Why won’t you let me through?”

He says, just as coldly, “I don’t know either.”

You explode. You start making calls, pulling connections, asking everyone you know why you are not being allowed through. But nobody tells you anything. You simply cannot get through. The immigration checkpoint begins to resemble Franz Kafka’s The Castle (《城堡》).

You want to enter the castle, but you never can. The only people you ever deal with are the lowest-level clerks in the castle. They do not know why. And even if they did, they would not tell you.

My motherland has an extraordinary literary quality. Even a new set of exit-and-entry regulations can become this damn Kafkaesque.

————————————————

“When China’s Iron Gate Closes, What Becomes of You?” (《中国的铁门关上，你的命运是什么》)

Whenever our country’s leaders become unhappy with another country, they can simply declare it a place of “the highest risk.” This may not be an objective description so much as rhetorical language. But rhetorical language can be far more destructive.

When I was a child in Xinjiang, there was a leader of the worker propaganda team named Old Liang (老梁). He could be especially destructive in this way. Often, when he denounced someone, he had not actually found any specific crime. The person might not have been a landlord, rich peasant, counterrevolutionary, bad element, or rightist. He might not have stolen corn from the commune or had an illicit affair. But at moments like this, Old Liang would leap onto a square table, point at the man, and shout:

“Him! His thinking is poisonous!”

And everyone would surge forward to denounce the poisonous man.

Nobody cared where the poison was, or whether it was snake venom or scorpion venom. What mattered was the infectious power of the word itself: “poisonous.”

If you accuse someone of a specific offense—say, “He stole a chicken”—people may start asking questions. How many chickens? Was it a rooster or a hen? But if you throw a vague label at someone—“bad element,” “bad character”—people can easily reach agreement inside a shapeless cloud of meaning.

Applied to an entire country, the phrase becomes something like an ideological nuclear bomb: “That country is high-risk. It’s poisonous.” Before long, even you begin wondering whether you should still go.

Of course, you may not be able to go even if you want to. Open your phone and look: flights to Japan have already been canceled in wave after wave, approaching the point of a complete shutdown.

Source: New Century (新世纪)

Wang Binbin

The concept of “unwritten rules,” introduced by Wu Si, has become extraordinarily popular. It helps people understand history and the countless strange events, large and small, that have taken place within it. But had the concept applied only to the past, it would never have gained such currency in the present. Wu Si distilled it from Chinese history. Yet within only a few years, it was already being used constantly by people from every walk of life. This shows that the concept speaks even more directly to contemporary reality.

The phrase was soon turned into a verb. Someone wants to subject someone else to the “unwritten rules”; someone has been “unwritten-ruled”; someone, in order to achieve a goal, allows themselves to be “unwritten-ruled.” Expressions of this kind appear every day on the internet and in newspapers and magazines.

In China, the unwritten rules often exert more power than the written ones. Even during the highly unusual period of the Cultural Revolution, did they ever really disappear?

The Cultural Revolution called itself “unprecedented in history.” In many respects, it was. It encouraged children to draw a line between themselves and their parents, and even incited them to assault their own fathers and mothers. Husbands and wives were urged to inform on each other and to compile reports on private remarks made in bed. Poverty was treated as a mark of honor. Calloused hands became a qualification for university admission, while handing in a blank examination paper could turn someone into a hero. Such phenomena were indeed without precedent.

Yet even in an age that claimed to be breaking completely with everything inherited from the past, many traditional practices survived with remarkable tenacity. Some continued under new names and altered appearances. Others persisted without changing either their names or their faces.

The unwritten rules continued to play an important role in social life during the Cultural Revolution, often in forms inherited directly from the past. Even this statement may understate their importance. The Cultural Revolution’s claim to be “unprecedented” should be understood in two ways.

On the one hand, many things happened that had never happened before. In that sense, the period was certainly unprecedented. On the other hand, many traditional practices emerged and flourished with an intensity never before seen. That too was unprecedented. During the Cultural Revolution, the unwritten rules were often more deeply entrenched than at any earlier time.

Sometimes these two forms of the “unprecedented” appeared in the same institution. University admission is one example. During the Cultural Revolution, the national college entrance examination was abolished. Universities recruited students directly from among workers, peasants, and soldiers. There were no academic examinations, and in some cases no meaningful educational requirements at all.

Perhaps the only firm requirement was that applicants be politically “red” and come from an acceptable class background. The children of landlords, rich peasants, counterrevolutionaries, “bad elements,” and rightists—as well as those themselves classified as class enemies—were not eligible to enter university. Never before had Chinese universities admitted students in this fashion. It was unprecedented.

Universities remained enormously attractive to young people during the Cultural Revolution. Whether one was a worker, a peasant, or a soldier, spending several years at a university as a “worker-peasant-soldier student” could lead to cadre status, a salary, and access to the state grain-supply system. Who would not want such an opportunity?

But there were very many politically reliable workers, peasants, and soldiers. Which of them would become the lucky few? At that point, the unwritten rules began to operate with unprecedented force.

Universities during the Cultural Revolution had no real authority to select students. They had only the power to receive those chosen for them. According to the formal rule—the rule written in documents and publicly proclaimed—workers, peasants, and soldiers were supposed to recommend candidates.

I lived in the countryside during the Cultural Revolution, so I do not know the precise situation among workers and soldiers. But in rural areas, the phrase “recommended by the poor and lower-middle peasants” was entirely empty.

University places were allocated to rural people’s communes. A particular commune might, for example, be allowed to send two students to university in a given year. Who those two people would be was decided by the commune’s highest official, the Party secretary. There was often not even a pretense of collective recommendation. Recommendation by all the poor and lower-middle peasants became recommendation by one man.

Whom did the secretary recommend? That depended on who was closest to him and whose gifts or bribes were of the highest quality. In determining who could attend university, the unwritten rules operated with unprecedented force. University admission during the Cultural Revolution therefore embodied both meanings of the phrase “unprecedented in history.”

The term “unwritten rules” now seems to have both a broad and a narrow meaning. The broad meaning needs no explanation. In its narrower sense, it often refers specifically to a young woman exchanging sexual access for something controlled by a powerful man. When people say that a certain “beauty” was subjected to the “unwritten rules” by a leader, everyone understands what is meant.

This narrow form of the unwritten rules was also widespread during the Cultural Revolution. Among the “worker-peasant-soldier students” admitted to university, those classified as peasants were often selected from the ranks of the “educated youth” living in the countryside.

These rural educated youth came from two groups: those who had returned to their home villages after schooling, and those who had been sent down from the cities. They could also, of course, be divided by sex.

A male educated youth who wanted to attend university often had to use money or gifts to clear the way. A female educated youth sometimes had to offer her body. Nor was university admission the only issue. Sent-down youth who wanted to be transferred back to the city, obtain permission to visit home, or simply avoid harassment and abuse had to pay a price. Men offered goods. Women sometimes had to offer themselves. No one knows how many female educated youth were sexually exploited by local officials during the Cultural Revolution. Many memoirs and works of fiction have described such experiences.

Today, the internet and tabloid press frequently discuss the unwritten rules of the film and entertainment industries. The most familiar example is the actress who must sleep with a director in order to obtain a role. Such things also occurred during the Cultural Revolution, perhaps to an even greater degree.

Dai Jiafang’s The Trials and Turmoil of the Model Operas, published by Knowledge Press in 1995, reveals something of Liu Qingtang’s conduct in this area.

Liu Qingtang played the Party representative Hong Changqing in the model ballet The Red Detachment of Women. He was a favorite of Jiang Qing and rose from the status of a performer to a position of considerable political power. He served as a delegate to the Ninth National Congress of the Chinese Communist Party and as a member of its presidium. At the Tenth Party Congress, he became a member of the Central Committee. He first entered the State Council Cultural Group and later became a vice minister of culture. Armed with high office and great influence, Liu committed every kind of wrongdoing.

Dai Jiafang describes Liu’s sexual exploitation of actresses in the following terms:“Once he gained power during the Cultural Revolution, his habitual promiscuity became shameless and completely unrestrained. A young woman more than twenty years his junior was kept by him for a long period. Any actress who wanted a leading role or a featured dance position, who hoped to secure household registration or employment for her husband, or who merely wished to avoid denunciation under conditions of terror might become his plaything if she needed something from him.

“Later, Liu Qingtang simply established his office beside the women’s bathhouse. In addition to a desk and filing cabinets, it contained a large bed, supposedly for resting while working late. He often waited by the door and watched for actresses returning from the bath. On various pretexts, he would invite them inside for a ‘talk,’ then use a mixture of enticement and coercion to force them into bed. Many actresses, whether vulnerable or simply terrified by his power, had no choice but to endure his abuse.”

Liu Qingtang’s immediate superiors, Jiang Qing and Zhang Chunqiao, were not unaware of this behavior. They knew about it but treated it as an insignificant personal failing, a “minor matter” that caused no damage at all to the career or standing of this sexual predator.

The opposite of an unwritten rule is a written rule: the formal regulations printed in documents, posted on walls, and constantly repeated in public. Written rules and formal procedures can at least place some restraint on hidden practices. During the Cultural Revolution, however, formal rules were widely weakened or simply abandoned. The power of the unwritten rules therefore became “unprecedented in history.”

The history of the twentieth century will record in its darkest pages two great human tragedies: fascist brutality and the catastrophe of the Cultural Revolution. Those who personally endured these calamities have tried to forget them, yet have found themselves unable to do so. The author of this book seeks to reveal the truth of that unprecedented disaster through an honest record of the inner experiences of ordinary Chinese people during the Cultural Revolution.

Source: Aisixiang

Chen Tushou（陈徒手）

He Unearthed Brutal Truths from the Archives

Chen Tushou copied hundreds of thousands of Chinese characters by hand in the Beijing Municipal Archives（北京市档案馆）. Some people would have preferred to see those files buried forever. In that sense, Chen was like a grave robber.

He spent years gathering the valuable things he had unearthed, then worked them into two books, What the People of the Old Country Thought（《故国人民有所思》） and When People Fall Ill, Does Heaven Know?（《人有病，天知否》）. The books examine the humiliations suffered by intellectuals during the “Thought Reform Campaign”（思想改造运动） in the early years of the People’s Republic and trace the ideological remolding of eleven university professors after 1949.

Huang Zongying（黄宗英） once called Chen and told him, “Feng Yidai（冯亦代） read your book and cried. The people you wrote about were all his acquaintances, but he knew nothing about these things. That was why he cried.”

During the nineteen-nineties, Chen copied hundreds of thousands of characters from documents in the archive of the Chinese Writers Association（中国作家协会）. “Most of them were meeting minutes,” he said, “along with letters from writers, statements of confession, and self-criticisms. Every political campaign left behind a pile of papers.”

During the Anti-Rightist Campaign（反右运动） of 1957, the Chinese Writers Association held more than twenty mass meetings to denounce the “Ding Ling–Chen Qixia Anti-Party Clique”（丁玲、陈企霞反党集团）. The meetings were often chaotic. People cursed, cried, and caused scenes. At one meeting, the Tianjin writer Liu Xi（柳溪） publicly accused Chen Qixia（陈企霞） of having had an “improper sexual relationship” with her. She even disclosed intimate details from the bedroom. The entire room was stunned.

When Chen Tushou copied the minutes of that meeting, he was shaken. “That was probably the most brutal meeting of all. These people had once been old friends, yet now they were fighting to the death in front of one another, using the most disgusting things they could find as weapons.”

To bring Chen Qixia down, Liu Baiyu（刘白羽）, Party secretary of the Writers Association, made a special trip to Tianjin to persuade Liu Xi to appear in person. Chen Qixia, then editor-in-chief of Literature and Art Gazette（《文艺报》）, had been “a proud and stubborn man.” He collapsed almost at once.

Chen summed up the Anti-Rightist Campaign this way: “It was one battle nested inside another, one person linked to another, as though everything had been arranged.” Ding Ling（丁玲） was attacked first, followed by Feng Xuefeng（冯雪峰）, and then the struggle returned to Ding Ling. At the final meeting, participants voted on whether to expel her from the Chinese Communist Party. Ding Ling herself raised her hand in favor.

The archival material used in When People Fall Ill, Does Heaven Know? amounted to less than half of what Chen had copied. Most of the information had never been made public.

In 2001, Chen “transferred” his operations to the Beijing Municipal Archives. Over time, the computer replaced pen and paper. Every historical source used in What the People of the Old Country Thought came from those archives. There was, he said, “no casual invention and no arbitrary expansion.”

Xie Xizhang（解玺璋）, critic and author of A Biography of Liang Qichao（《梁启超传》）, offered this assessment: Chen Tushou’s studies of contemporary Chinese intellectuals are consistently grounded in large quantities of official material and original archives. Not a single sentence lacks a source.

Digging Through Archives by Hand for More Than a Decade

Born in 1961, Chen describes himself as a member of “a politicized generation.”

He was in junior and senior high school during the Cultural Revolution（文化大革命） and had very little time for ordinary study. Students attended a few classes in the morning. In the afternoon, everyone went into the streets to read big-character posters（大字报）. The posters were pasted so high that people had to climb ladders to put them up. Huge black characters covered sheets of red paper. Chen would stand below and look up. After a while, under the sunlight, the red paper seemed to turn into a solid black mass.

Every morning, the teacher selected five or six students from a class of more than forty and asked them to repeat what they had read. Chen was once selected. He could not fully understand the posters. He knew only that one person or another was being denounced. He repeated the words one by one: Party secretary Liao Zhigao（廖志高） was practicing revisionism.

Later, the struggles and “movements” described on the posters became part of his own lived experience. In 1986, Chen was transferred from the Central Committee of the China Zhi Gong Party（中国致公党中央） to the Creative Liaison Department of the Chinese Writers Association. He discovered that “although the effort to restore order after the Cultural Revolution had been underway for years, the Writers Association was still extremely complicated. It had never really known peace.”

In January 1987, the Party launched a campaign against “bourgeois liberalization”（资产阶级自由化）. The Writers Association also organized repeated study sessions and political campaigns. At each council meeting, representatives of the left and right rushed to speak, accusing the other side of promoting liberalization. The atmosphere was explosive.

At one congress of writers and artists, Chen took the minutes. Wang Meng（王蒙） said, “There is a problem here. You should not keep sounding the alarm to the central authorities whenever anything happens.”

“That shows that some people were constantly filing complaints,” Chen later recalled.

That winter, the Writers Association held a conference for young writers at the Jinfeng Hotel（金丰宾馆）, then located in Beijing’s western outskirts. During the day, participants studied literary theory—socialism and realism. At night, they watched foreign films. The meeting continued for eight or nine days. Snow fell outside, and the sky remained dim.

During a rectification campaign at the Writers Association in the late nineteen-eighties, the secretary and deputy secretaries were removed. Every member of the Party leadership group was replaced. Liu Baiyu, already in his seventies, returned to the association and became editor-in-chief of People’s Literature（《人民文学》）.

The atmosphere of the nineteen-fifties, when “everyone had to pass inspection,” returned. After the political storm, work at the association came to a halt. Staff members had little to do but play chess and cards every day.

The archive room stood next to Chen’s office. Documents brought back from cadre schools（干校） were piled in burlap sacks against the wall.

Colleagues in the archive asked Chen to help organize the files. Once the work was finished, he took documents home and copied them. He continued transcribing these “confidential” institutional archives until he left the Writers Association in 1993.

Outwitting the Archives

Around 2003, Chen Tushou and the historian Yang Kuisong（杨奎松） were both frequent visitors to the Beijing Municipal Archives. A few years later, Yang wrote about the transformation of Wang Yunsheng（王芸生） and Ta Kung Pao（《大公报》） after 1949. Chen began publishing articles in Idle Talk from the Central Hall（《中堂闲话》） and Yanhuang Chunqiu（《炎黄春秋》）, describing how professors at leading universities, including Yu Pingbo（俞平伯） and Feng Youlan（冯友兰）, underwent ideological remolding after 1949.

Archival management was inconsistent, and gaining access often depended on luck.

After interviewing Wang Lang（汪朗）, the son of the writer Wang Zengqi（汪曾祺）, Chen still needed the institutional side of the story. He decided to visit the Beijing Peking Opera Troupe（北京京剧团） and look at Wang Zengqi’s personnel file.

On the day he arrived, a retired member of the troupe had just died, and the senior administrators had all gone to handle the funeral arrangements. Only two young women remained. When they heard that Chen was conducting an interview, they were delighted. One of them pointed and said, “Go ahead and look.”

A heap of battered documents had been tied together with coarse hemp rope. The three characters “Wang Zengqi” were written across the top in brush and ink. The paper inside was poor in quality and irregular in size. There were notebooks, manuscript paper, and documents typed in triplicate using carbon paper. Chen felt as though he had discovered buried treasure.

Once he began copying, the head of the archive office could do little about it and eventually left him alone. The files contained more than one hundred thousand Chinese characters. Chen copied them two or three times, but he could still use only part of the material. When People Fall Ill, Does Heaven Know? quotes Wang Zengqi’s post–Cultural Revolution self-criticisms and other documents, all of which were entirely new and had never been published.

Afterward, at Chen’s suggestion, Wang Lang asked the Peking Opera Troupe to give him the files so that they would not be destroyed. The archive office refused. Family members, it said, were not permitted to see them.

The archives lacked a systematic classification scheme. File titles often had little to do with their actual contents. Simply reading the catalogue would never lead a researcher to the material he wanted.

For a time, Chen studied the architect Liang Sicheng（梁思成）. He wanted to find Liang’s speeches at sessions of the Beijing Municipal People’s Congress during the nineteen-sixties. The catalogue yielded nothing. The archive’s method had been to bind all the speeches delivered at People’s Congress meetings into thick volumes. Chen had no choice but to read them from beginning to end. At last, he found several of Liang’s speeches.

Chen had read nearly every commercially available biography of Liang Sicheng. The books were attractively arranged and sold well, but they “copied one another and offered nothing new.” In the end, they all reinforced the same account of “how Liang Sicheng followed the socialist path after Liberation.”

In archives and oral testimony, Chen discovered another side of Liang, one that had never been presented to the public.

During the Anti-Rightist Campaign, Liang Sicheng was harshly criticized and thoroughly humiliated. Later, he too began attacking other people. At a meeting held to denounce the scientist Qian Weichang（钱伟长）, Liang spoke fiercely: “We will use our T-squares to beat you into meat paste.”

“Beat you into meat paste” was one of the expressions commonly used by workers, peasants, and soldiers during the Anti-Rightist Campaign.

In an Absurd Age, No One Remained Clean

Not long ago, Wang Keming（王克明） sent Chen an email. He had read the chapter on Feng Youlan in What the People of the Old Country Thought and found it “fairly balanced.” Feng Youlan was Wang Keming’s uncle by marriage.

Zi Zhongyun（资中筠） also passed along a similar assessment from Feng Youlan’s daughter, Feng Zhongpu（冯钟璞）. “Feng Zhongpu has always had very high standards. She rarely approves of other people’s writing.”

Chen’s judgment of Feng Youlan is this: “His experience has the quality of a historical specimen. He was constantly involved in political struggles, yet he also remained slightly detached. At critical moments, he knew how to withdraw. He would always catch his breath and somehow survive again.”

Chen understands Feng’s participation late in life in the “Liang Xiao Writing Group”（梁效写作组） in much the same way: “All he did was revise annotations to classical Chinese texts for other people. It was hardly a great crime, yet people later described it in the most disgusting terms.”

Even someone as decent as Lao She（老舍） instinctively tried to avoid danger.

“When the ‘rightist’ Wu Zuguang（吴祖光） was being denounced, Lao She frequently spoke at meetings and echoed the accusations. But in private, he took good care of Wu. When Wu Zuguang was later sent to the Great Northern Wilderness（北大荒） for labor reform, Lao She bought some of Wu’s paintings and gave them to Wu’s wife, Xin Fengxia（新凤霞）. That was how he expressed his concern for them. When Wu later wrote about it, he said with emotion that Mr. Lao She had still been a very good man,” Chen told a reporter.

“In 1964, during the rectification campaign in cultural circles, the Beijing Municipal Party Committee wanted to protect itself, so it pushed Lao She forward as a target. In August 1966, the Red Terror was especially severe. Yet the situation eased somewhat after that month. If Lao She had not taken his own life, he might have endured it.”

Chen explains Lao She’s suicide this way: “After 1949, he had never really suffered much. When this suddenly happened to him, he could not bear it.”

“We cannot judge them too harshly. The age itself was absurd.” Intellectuals as a group shrank under the pressure of those extraordinary years, but Chen still wants to show their capacity for goodness and decency.

“Take Shen Congwen（沈从文） and his A Study of Ancient Chinese Costume（《中国古代服饰研究》）. He wrote that book entirely out of a sense of respect for the country. He studied silk and handled the objects one by one. That is why the book has such weight. I deeply respect that kind of professional devotion.

“And there was Mr. Lao She. He wrote one play after another. Each would be performed a few times, and once the Great Leap Forward（大跃进） had passed, the play would disappear with it. Yet he maintained his enthusiasm for writing. It reflected the intense hopes that intellectuals of that generation once placed in the new era.”

Chen often asks himself what he would have done in their place. “Perhaps I would have behaved even worse. If the Gang of Four（四人帮） had not been overthrown, and if Reform and Opening（改革开放） had not begun, all of us who came afterward might have suffered the same fate.”

“You had to protect yourself at all costs. That is the most humiliating and heartbreaking part of the Chinese intellectual’s experience. But in that political environment, anyone might have been forced to dirty his hands. If your hands did not become dirty this time, they might the next. No one came through completely clean,” Chen said.

Source: New Three Cohorts（新三届）

Li Zhide

On July 29, 2026, The Saturday Evening Post published a column by Ben Railton, a scholar of American history and American studies, titled “Does Remembering Historical Injustice Erase America’s True Story—or Help Us Better Understand It?” The article responded to a 162-page report issued by the White House Domestic Policy Council on July 4, Saving America’s Story: How the Ideological Capture of the Smithsonian’s National Museum of American History Erases Our Heritage. The White House report singled out “Many Voices, One Nation,” an exhibition at the Smithsonian’s National Museum of American History, for particular criticism. The exhibition explains that nineteenth-century white Americans felt threatened by the racial identity of Chinese immigrants, helping to produce the Chinese Exclusion Act of 1882 and a succession of restrictive immigration laws. The White House report accused the exhibition of portraying those laws as “tools of white supremacy.”

Railton’s response is more than a simple moral denunciation. He reconstructs the policies, rhetoric, and violence that preceded the Chinese Exclusion Act, showing that exclusion was not a sudden legislative accident. It had a long period of preparation. In 1850, California passed the Foreign Miners’ Tax Act, imposing a monthly tax of twenty dollars on foreign miners, including Chinese and Latin American workers. After that law was repealed, California introduced a new tax of four dollars a month in 1852, with tax collectors directing most of their efforts at Chinese miners. Long before federal law formally excluded Chinese immigrants, local government had already defined them as foreigners who could be treated differently.

The Page Act of 1875 carried this local discrimination into federal law. Drafted by California congressman Horace Page, it is often regarded as the first federal immigration law designed to create a systematic structure of exclusion. In the name of combating contract labor, prostitution, and supposedly “immoral activities,” the law subjected immigrants arriving from Asia to special inspection. It did not formally declare that all Chinese women were to be excluded. In practice, however, enforcement focused overwhelmingly on them. Large numbers of unmarried Chinese women, or women traveling without male companions, were presumed to be prostitutes and forced to prove their moral character through severe and intrusive questioning. Chinese men could still be admitted as cheap labor, but they found it extremely difficult to bring wives and families into the country. From the beginning, Chinese exclusion was more than a labor policy. It was also a system designed to obstruct family formation and prevent Chinese communities from reproducing themselves.

Street violence came before the federal laws. On October 24, 1871, a mob attacked the Chinese community in Los Angeles, looting homes and businesses and lynching nineteen Chinese residents. The victims made up more than one-tenth of the city’s Chinese population of 172. Several members of the mob were convicted, but the California Supreme Court overturned the verdicts in 1873. In the end, not one of the attackers served a prison sentence. What made the episode so terrifying was not only the number of people killed. It showed that when racial violence received judicial tolerance, the mob learned that some lives did not enjoy the same protection as others.

In 1877, economic depression and unemployment pushed labor unrest in San Francisco toward a crisis. After a rally at the Sand Lot, the violence quickly turned against the Chinese community. Chinese-owned laundries, shops, and homes were attacked and burned. Denis Kearney later organized this anger into a more durable political movement. He compressed complicated economic problems into a single slogan: “The Chinese must go.” Railroad capital, falling wages, unemployment, and widening inequality were pushed aside. Chinese immigrants became the most convenient scapegoats. Kearney later acknowledged that his aim was to turn local anti-Chinese sentiment into a national political issue and ultimately to stop Chinese immigration.

This language was not confined to the streets. In his 1875 State of the Union address, President Ulysses S. Grant described Chinese women as an “evil” requiring congressional action. John Marshall Harlan, the Supreme Court justice later celebrated for defending equal rights for Black Americans, also described the Chinese as “a race utterly foreign to our own” that would never assimilate. The remark did not appear in a majority opinion formally written by Harlan. It was connected to his discussion of exclusion policy in relation to the Wong Chin Foo case. He also joined the dissent opposing the recognition of birthright citizenship for American-born children of Chinese parents. This detail matters. A person who opposes discrimination against one racial group is not necessarily free from the prejudices of his age when confronting another.

Railton therefore raises a question that cannot be avoided. When a nation tells its own history, does acknowledging violence, prejudice, and institutional exclusion amount to defaming the country? Or does it explain how the country departed, step by step, from its own principles? “Many Voices, One Nation” does not claim that American history consists only of oppression. Nor does it deny the American system’s capacity for reform. It simply shows that the country’s traditions of freedom and exclusion have always existed side by side. Chinese immigrants helped build the railroads, worked in the mines, and sustained urban economies, yet the law declared them an alien race incapable of assimilation. This contradiction does not damage the American story. It is part of the American story. The Smithsonian’s own description of the exhibition has consistently emphasized that it seeks to explain how the United States became one nation out of different populations, migration experiences, and cultural traditions.

In earlier discussions, it was still possible to say that defending the historical background of the Chinese Exclusion Act might lead to political action in the present. That judgment now needs to be updated. On July 24, President Donald Trump signed Executive Order 14416, formally turning the conclusions of Saving America’s Story into administrative policy. The order instructed the secretary of the interior, the director of the Office of Management and Budget, the administrator of the General Services Administration, and other officials to use their existing authority, funding conditions, and contracting powers to compel the Smithsonian to “correct” the problems identified in the report. It also directed the National Park Service to place temporary signs outside the National Museum of American History, informing visitors that the museum’s exhibitions reflected alleged “ideological capture” and providing what the government defined as “accurate information.”

This does not mean that the United States has reenacted the Chinese Exclusion Act. Exaggerating that point would weaken the credibility of the criticism. But this is no longer merely an academic dispute. The government has begun deciding how museums should describe history, which interpretations qualify as “accurate,” and which should be condemned as “ideological manipulation.” It is also preparing to use administrative authority and financial pressure to influence the content of exhibitions. Only twenty days passed between the report and the executive order.

This is how historical narrative becomes connected to political power. A state does not usually begin by announcing that it intends to restore a racial law already widely regarded as a national disgrace. It begins by changing the standard by which that law is judged. Discrimination is redescribed as the maintenance of order. Racial exclusion is presented as a practical necessity. The memory of victims is dismissed as ideological propaganda. Once society grows accustomed to this language, new forms of exclusion no longer need to carry the name of an old law. They need only new targets, new justifications, and new administrative instruments.

The real danger of the White House report, then, lies in more than its interpretation of the Chinese Exclusion Act of 1882. More troubling is the historical principle it seeks to establish: whenever a government invokes national security, cultural conflict, competition for jobs, or pressure from immigration, dividing people by race or national origin may be treated as an understandable, perhaps even defensible, “practical choice.”

Once that premise is accepted, the next step will not remain inside the museum. Today, officials may demand that an exhibition remove the language of “white supremacy.” Tomorrow, collective suspicion directed at an immigrant group may be justified as national security. More severe identity checks may be presented as administrative efficiency. Unequal treatment may be described as the protection of tradition. History does not return in precisely the same form. It changes its language and puts on a new uniform. It reappears through new forms, databases, budget provisions, and executive orders.

This is also the value of Railton’s article. He does not treat the Chinese Exclusion Act as an artifact sealed inside the nineteenth century. He reveals a complete political chain: economic anxiety is rewritten as racial resentment; racial resentment becomes a public slogan; public slogans encourage local violence; local violence receives judicial tolerance; political leaders then write prejudice into federal law. The present may not repeat every link in that chain. But denying historical injustice removes the first line of defense against its return.

Li Zhide

A troubling shift is taking place in immigration policy during Donald Trump’s second term. In the past, the government mainly described its deportation targets as people who had crossed the border illegally or had criminal records. Now the policy is turning toward a different group: people who entered through government-created programs, registered with the authorities, underwent background checks, received work permits, paid taxes, and lived according to the rules. The government once allowed them to remain. It is now withdrawing those protections through administrative action.

According to estimates by the Migration Policy Institute, during the first year of Trump’s second term, temporary protections were revoked for more than 1.5 million people admitted under humanitarian parole programs. As of March 2025, nearly 1.3 million people in the United States held Temporary Protected Status, or TPS. Some of these programs overlap, and the dates on which protections were terminated vary. Even so, large-scale “de-legalization” has clearly taken place.

The term “de-legalization” captures the heart of the problem. These people did not secretly cross the border overnight, nor were their documents suddenly exposed as fraudulent. Yesterday they were legally permitted to work. Today they may lose both their work authorization and their protection from deportation. The state first creates a lawful status, allows people to build careers, start families, and open businesses under that status, and then removes it by executive decision. The undocumented population is therefore no longer produced only at the border. It can also be created by the government itself.

Temporary Protected Status and humanitarian parole are, of course, temporary legal arrangements. They are not green cards, much less citizenship. The Trump administration’s argument is also clear: temporary programs cannot be renewed indefinitely, or they become permanent residence in all but name; immigration policy falls within executive authority and powers granted by Congress; and courts should not substitute their judgment for the government’s assessment of whether conditions in another country remain dangerous. On June 25, 2026, the Supreme Court ruled six to three that the Trump administration could terminate temporary protections for roughly 350,000 Haitians and more than 6,000 Syrians. The Court also held that the relevant law sharply limits judicial review of decisions made by the secretary of homeland security.

That legal conclusion answers the question of whether the government possesses such authority. It does not answer the question of how that authority should be used. Calling a program “temporary” does not allow the government to ignore the fact that many people have lived in the United States for ten, fifteen, or even more years. Many Haitian TPS holders received protection after the 2010 earthquake. Since then, Haiti has endured the assassination of its president, the collapse of state institutions, the expansion of armed gangs, and widespread violence. The U.S. State Department still maintains its highest-level travel warning for Haiti. The government tells American citizens not to go there while telling Haitians that the country is now safe enough for them to return. That contradiction cannot be explained away simply by repeating that temporary status is temporary.

The question of whether the United States is returning to a white nation is therefore not an entirely emotional or baseless one. Early American institutions did contain an explicit structure of white national preference. The Naturalization Act of 1790 limited eligibility for naturalized citizenship to “free white persons.” The Chinese Exclusion Act of 1882 was the first federal law to systematically bar a group on ethnic grounds, preventing Chinese laborers from entering and denying Chinese residents a path to citizenship for decades. The Immigration Act of 1924 created national-origins quotas that favored immigrants from Western and Northern Europe while largely excluding Asians. The United States did not abolish this openly racial hierarchy of immigration quotas until 1965.

The United States has not restored the racial quotas of 1924, and the countries from which immigrants come are not, by themselves, enough to prove that the government is building a legally defined white nation. The reality is more complicated. Some of the programs terminated by the Trump administration also affect citizens of European and Middle Eastern countries. Supporters of the policy are concerned not only with race, but also with border control, presidential authority, competition in the labor market, and the transformation of temporary administrative programs into long-term arrangements. Reducing every immigration policy to white supremacy would obscure the disorder that has accumulated within the immigration system itself. It would also let Congress escape responsibility for refusing, year after year, to reform the law.

Race, however, cannot be removed from the discussion. In her dissent in the Haitian TPS case, Justice Elena Kagan noted that the plaintiffs had submitted evidence that included Trump’s racially charged statements. In her view, racial considerations had entered the decision to terminate the protections. The majority rejected that conclusion, but it did so largely on the ground that the law limited judicial review. It did not establish that racial motives were entirely absent from the policy.

The practical effects deserve even closer attention. Immigrants from Haiti, Venezuela, Cuba, Nicaragua, and other countries are heavily represented in health care, food service, hotels, construction, and small businesses. When temporary protections are abruptly terminated, employers must dismiss workers who had previously been legally employed. Families lose income, older people lose caregivers, and local governments lose taxpayers. The Associated Press recently reported that about one-third of home health aides in the United States are immigrants, and some elder-care facilities already fear that they will be unable to replace Haitian workers who are forced to leave.

The wholesale revocation of legal status may also create a vast “shadow population.” People who know that returning home could expose them to violence, persecution, or economic ruin will not simply disappear. They are more likely to move into the underground labor market. They may become afraid to report crimes, seek medical care, or send their children to school. They will also become easier for employers to exploit. On paper, the government may reduce the number of “legal immigrants.” In practice, it may increase the undocumented population, expand the supply of cheap labor, and enlarge the underground economy. The law pushes people into the shadows and then uses their life in the shadows as a reason to intensify enforcement. It is a self-reinforcing cycle.

The real choice facing the United States is not between an “open border” and a “white nation.” A country plainly has the right to control its borders and to set limits on temporary status. But order must include predictability. When a government invites people to enter through lawful procedures and allows them to remain for years, it assumes a degree of public responsibility. Status may be revised, but procedures cannot be arbitrary. Policy may become stricter, but people should not be transformed overnight from law-abiding residents into deportation targets.

Whether the United States is becoming a white nation again cannot be judged only by the percentage of the population that is white. A more important measure is how the state decides who deserves to remain, whose life may be erased by administrative power, who is treated as a member of the nation, and who remains merely a source of labor to be used when needed and discarded when convenient.

In 1924, the United States used national-origins quotas to preserve a white majority. Today’s policies may not copy those laws directly, but they may restore their deeper logic: ranking immigrants by country of origin, skin color, culture, and political imagination. The law no longer openly declares “white people first.” Yet selective termination of legal protections can produce a similar social result.

The United States has not yet returned to that white nation. Courts, state governments, civil society organizations, businesses, and a multiracial public are still resisting this retreat. But the danger is already visible. When the lawful lives of more than a million people can be rapidly revoked, when the government deliberately creates an undocumented population, and when belonging depends entirely on a single administrative decision, the United States is losing more than immigrants.

It is losing one of its most important political promises: that a person’s rights should be protected by stable law, not determined by whether those in power decide that person still counts as one of “us.”