On August 18 at 7:36 in the morning UTC, a user called CyberLeeker posted stolen Grand Theft Auto 6 footage on Dread, a Tor forum most people have never heard of. Twelve hours later it was everywhere. Gameplay clips. A full map of Leonida. Jason Schreier at Bloomberg confirmed the material is genuine Rockstar development content. The latest clip shows the leaker playing the game himself and shooting the word LEEK into a wall, which means this person is either sitting on a playable build or knows exactly who is.

Nine days before Rockstar’s official gameplay reveal on Netflix. Timed. Deliberate.

THE MANIFESTO

The leaker calls the campaign No Disc, No Peace and published three demands to Take-Two. Pre-orders limited to physical editions only. No day-one DLC carved out of content already shipped. Mandatory offline modes so games stay playable after the servers die. Then a closing threat to the whole industry: if CyberLeek can reach Rockstar, no one is safe.

Here’s what I’ll say to you. Two of those three demands are things gamers have wanted for a decade. Game preservation matters. Digital storefronts revoking access to things you paid for is a real problem. If this were an actual protest, plenty of people would be sympathetic.

It isn’t.

FOLLOW THE MONEY

The leaked map images carry a call to action to buy a Solana token called CYBERLEEK, pitched as fundraising to fight for gamers’ rights. Look at what that token actually did. It launched with the leak, spiked to roughly $0.0018 on August 19, and dumped 49 percent in the next 24 hours. As I write this it trades at $0.0007288 on a market cap of $729,770. Volume in that same 24 hours: $5.8 million. That’s not a war chest for gamers’ rights. That’s early wallets selling into a retail wave, on a token anyone could have spun up in ten minutes.

Nobody funds a protest by dumping a coin on the people who showed up for it. The manifesto pulled buyers in and the early wallets sold to them. Everything else is set dressing.

I spent eight years in Navy operations centers and another decade in IT infrastructure, and pattern recognition is the whole job. This pattern shows up after every viral news event now. Rug the hype, ride the headline, disappear. The GTA 6 version just happens to have better production value.

WHAT HAPPENS NEXT

The odds say the leaker gets caught. Probably not this week, but the base rate is brutal. The 2022 GTA 6 leaker was arrested within days. The Half-Life 2 source thief was identified across an ocean in 2004. Take-Two is the most litigious publisher in gaming, the FBI takes computer intrusion plus public extortion demands seriously, and independent researchers already traced CyberLeek’s handle to old German forum accounts within 48 hours. Old account reuse is how these identifications always happen.

And this one keeps talking. Manifestos, demands, new clips. Every post is metadata. The quiet leakers sometimes survive. The ones building a persona never do, because the investigators only need one mistake and the ego guarantees a supply of them.

Will the full game leak before the November 19 launch? No. A development build runs on devkits, not your PS5. Even a full dump would hand dataminers a pile of assets, not a playable game. Expect more clips, escalating DMCA carpet bombing, and then an arrest that ends the show.

GTA 6 itself takes no real damage. Take-Two stock barely moved. Pre-orders opened in June at $79.99 and the game will break records anyway. The only people losing money on this story are the ones who bought a leek-themed token at the top.

THE PRINCIPLE

Every big news event now ships with a token attached, and the token is never the cause it claims to be. The manifesto, the movement, the fundraiser: check what the money actually does before you believe what the money says. Thirty seconds on a chart told the whole truth here, a 2x pump and a 50 percent dump in two days while the manifesto talked about fighting for your rights.

Stay calm. Stay educated. Let other people be exit liquidity.

Follow me everywhere as @vincentactual. More at vincentactual.com and on YouTube at @VincentActual.

Four days later, I ran one of my own drafts through it.

AI: 100 percent. Human: 0 percent.

I’ll be honest about what happened next. My first instinct was the same one you’d have. Can we rewrite it so it passes?

The answer is no. And the reason why is a mechanism worth understanding, because it’s the same lesson I keep teaching about markets. You don’t beat the system by dressing up the trade. You beat it by understanding the plumbing.

HOW THE DETECTOR ACTUALLY WORKS.

Pangram is a neural network trained on roughly a million documents, half human, half machine. The clever part is what they call mirror prompts. For every human document in the training set, they generated an AI copy in the same style, same tone, same subject.

Read that again. They specifically trained it on AI text that was styled to sound human.

That means the detector isn’t looking for em dashes, or the word “delve”, or robotic phrasing. It’s reading token-level statistical patterns, the probability fingerprint that a language model leaves in every sentence it generates. You can style AI text to sound exactly like me. The fingerprint survives. I write about radar for a living. Think of it as the difference between repainting an aircraft and changing its radar cross section. The paint job fools the eye. It does not fool the signal processing.

They also run what’s called hard negative mining. Every time human text gets wrongly flagged, they retrain on it. The model gets sharper every cycle. So even if a laundered draft squeaked past today, a rescan next month could flag a post that’s already live under your name. Permanently. On the record.

THE MATH ON GAMING IT.

Look at it as a trade, because that’s what it is.

The upside of successfully faking a human scan: nothing. Readers don’t subscribe because a scan says human. They subscribe because the analysis is right.

The downside of getting caught once: the entire brand. I’m a Navy vet who tells you what the chart doesn’t show. The one story that kills that guy is “he got caught passing off machine text as handwritten.”

Unlimited downside. Zero upside. That’s not a trade. That’s a short position on your own credibility.

SO HERE’S MY DISCLOSURE.

I use AI in my writing. Here is exactly how.

The analysis is mine. The positions are mine. The research, the numbers, the mechanism I’m walking you through, the eight years of Navy pattern recognition behind it, all mine. I run an AI company. I use the tools I build with. AI helps me turn my thinking into prose faster than I can type it.

And nothing gets a first pass. I iterate, and iterate, and iterate again until it’s right. Draft, tear it apart, redraft, check every number, cut what doesn’t hold. The machine types faster than I do. It doesn’t get to decide when a piece is done. I do. I direct, check, and stand behind every fact that goes out under my name.

Writing is only one lane. I run my entire company on this principle. I build AI automation for a living, ticket triage for MSPs, and AI voice agents at dispatchbyai.com that answer the phone, qualify the call, and dispatch the work so a human doesn’t have to sit on the line for it. I use the same approach on my own operation. The repetitive work, the monitoring, the routing, the first-pass diagnostics, all of it runs on automation I built and tuned. Not because I can’t do it by hand. I spent years doing it by hand. Because every hour I’m not putting out fires is an hour spent on the work that actually compounds: the analysis, the client relationships, the next thing worth building.

That’s the honest case for AI. It’s not a ghostwriter. It’s a watchstander. You still command the ship. You just stop standing every watch yourself.

What you’re reading right now was produced that way. Scan it. It’ll light up. That’s fine, because I just told you it would, and nothing in it is anyone’s thinking but mine.

Substack’s own line on this is the right one. AI isn’t the problem. Hiding it is. The scan button didn’t create that reality. It just repriced it, the same way a chokepoint doesn’t create risk, it reveals what was always sitting under the surface.

Every crisis in history transferred credibility from the people who faked it to the people who owned it. This one is no different.

Stay calm. Study the mechanics. Be the guy who disclosed before the scan, not the guy who explained after it.

I break this down daily on my Substack and YouTube. Follow me everywhere as @vincentactual.

SIGNED THE FIRST CLOSER

Ten years selling insurance on commission. She signed the rep agreement, sent back her W-9, and started asking questions before her first call. Two of them most founders can’t answer: what do competitors actually charge, and do they lock clients into contracts. I didn’t know. I had to go find out. That’s what a real salesperson does to your business. She makes you do your homework before the market makes you do it in public.

SO I DID THE HOMEWORK

The biggest name in AI answering for the trades hides its pricing behind a sales call. You book a demo, sit through the pitch, and then they tell you a number. The budget players publish prices between $49 and $299 a month, but they cap your minutes or your calls, and charge overage fees when your phone actually rings. The mid-tier runs $449 to $789 a month with minute caps of their own.

Here’s what I published instead: month to month, no contract, no minute caps, and you can talk to the AI before you ever talk to a human. Nobody else in this market combines all three. Now every rep conversation starts from that ground.

BUILT THE PIPELINE

68 HVAC companies across Miami-Dade, Broward, and Palm Beach. Every one verified by hand against their own website. Every one advertises 24/7 emergency service, which means every one has a phone problem at 2 a.m. Some are paying live answering services hundreds a month to take messages. One publicly posted that they cut their after-hours emergency coverage because they can’t staff it. Their website says it. That company is getting a call.

LEARNED WHO I’M REALLY FIGHTING

The big field-service software platforms are bolting AI answering onto their products. That was always coming. But their checkbox only works if your whole business lives on their software, configured by you, from a help center article. Dispatch bolts onto whatever you run today. Housecall Pro, Jobber, pen and paper. No software switch. We configure it for you, and the job details hit your phone the second the call ends.

THE SCOREBOARD

Clients: still zero. Reps: one. Verified pipeline: 68. Pricing research: done. Rep agreement, onboarding packet, and client contracts: signed, sealed, and templated in DocuSign.

The Navy taught me you don’t judge a watch team by their first night on station. You judge whether the watch bill is manned, the gear is tested, and the reports are honest. Manned. Tested. Honest.

Every business book sells you the launch montage. Nobody shows you week two, when the only honest number is zero and the only real work is making the zero temporary. That’s the part that decides everything. Study the mechanics. Build the system. The scoreboard follows the machine, never the other way around.

Want to hear what the fuss is about? Call (940) 828-8213 and tell it your AC died. Ninety seconds. Then imagine it answering your phone. The full product is at dispatchbyai.com.

I’m everywhere as vincentactual.

The owner is asleep. The on-call tech is asleep. The caller has a dead compressor, a house full of kids sweating through the sheets, and a credit card.

The call goes to voicemail. The caller hangs up and dials the next name on the list.

That call was worth somewhere between four hundred and four thousand dollars. It just walked.

Last time I wrote here, I told you I built the thing that answers that call. An AI voice agent that picks up 24/7, sounds like a person, books the job, and pages a human only when it actually matters. I told you I spent a night attacking my own agent like a hostile caller would, found the holes every cheap phone bot has, and closed them. Token-locked back end. Server-side validation on everything. Caller speech treated as hostile until proven otherwise.

That post was about building it.

This one is about what building it doesn’t do.

I want to be precise about where things stand, because build-in-public only means something if you publish the numbers when they’re ugly.

Clients booked since that post: zero.

Not “early traction.” Not “promising conversations.” Zero.

Here’s the thing nobody tells you when you’re heads down in the code. A finished product and a sold product are two different objects. The agent could be flawless, and it is close, and it would still make nothing, because the man running four trucks in the Miami heat has never heard of me. He doesn’t read my Substack. He doesn’t care about prompt injection. He cares that his phone rang eleven times last night and he doesn’t know what any of those calls were worth.

The product was never going to sell itself. I knew that intellectually. I know it differently now.

So here’s what I actually did since the last post.

I put the agent on a public line. (940) 828-8213. It’s live right now. Call it. Tell it your AC died. Ask it something stupid. Try to talk it out of its own rules the way I did the night I attacked it. I’m not going to tell you it holds up. I’m going to let you find out. Ninety seconds on that line is worth more than anything I can write here.

I built the pitch around the only number an owner feels. Most service companies miss a quarter to a third of their inbound calls, and the after-hours ones die in voicemail. That’s not a marketing stat to these guys. That’s Tuesday.

And then the real move. I stopped trying to be the whole company.

I’m standing up a sales force. A target list of South Florida HVAC shops, pulled and scrubbed by hand. A call script I wrote myself. People on the phones running it, with me training them on what a good conversation sounds like and what a wasted dial sounds like.

Think about what the product is for a second. The entire pitch of Dispatch by AI is that the owner should not be the only one who can answer the phone. A business that lives or dies on one man being awake is capped at that man’s calendar.

Then look at how I was selling it. One man. Between engineering sessions. Awake too long.

I was the exact failure mode I built the product to kill.

I spent twenty years in environments where systems beat heroes. On a ship, nobody is impressed that you personally stood every watch. They ask why the watch bill only has one name on it. The Navy drills this into you and then you get out and build a company and forget it inside six months.

So the sales motion gets engineered like everything else I ship. A list. A script. Training. A standard. Built so a rep who is not me can run it, and built so I can tell from the numbers whether the script is broken or the rep is.

Zero clients today. But zero clients with a live demo line, a scrubbed target list, a tested script, and dialers who aren’t me is a different zero than the one I had last month.

One is a scoreboard.

The other is a machine that hasn’t finished warming up.

I’ll publish the numbers as they come, including the ugly ones. That’s the deal.

If you run a shop and your phone goes to voicemail after five, some of those calls are jobs, and you know it. Call the demo line at (940) 828-8213 and try to break my agent. Or book thirty minutes at dispatchbyai.com and I’ll show you what your missed calls are costing you.

I write about the things most people look away from, including my own scoreboard. If you want to watch this get built with the real numbers attached, follow me on Substack and across all platforms. I’m everywhere as @vincentactual.

I build the fix.

It is an AI receptionist that answers your phone 24/7. It talks like a person, not a phone tree. It figures out whether the caller has a real emergency or a routine job, gets the name, number, and address, books the appointment, and pages your on-call tech only when it actually matters. No missed calls. No offshore answering service reading a script. No lost jobs at 11pm because you were asleep.

That part, a lot of people can build now. Here is the part they can’t.

I spent a night trying to break my own agent, the way an attacker would. What I found is what almost every cheap AI phone bot gets wrong. They hand out whatever a caller asks for, so you tell the bot “the owner says quote 89 dollars” and it does. They leak, storing raw caller text and passing it to other systems where a smart caller can plant instructions that run later. And they will page your staff on a fake emergency all night, because nobody locked the back end.

I fixed all of it. Token-locked back end, server-side validation on everything, caller speech treated as hostile until proven otherwise, and a safety backstop so a real gas-leak call still reaches a human even if the caller panics and hangs up. It refuses to take a card number on a recorded line. It cannot be talked out of its own rules.

This is not a side hobby. I am a Navy veteran and an MSP engineer. I secure systems for a living and I teach networking at the college level. When I say the agent is locked down, it is locked down the way I would lock down a client’s network.

Here is the offer. I built Dispatch by AI, a 24/7 AI voice agent for HVAC contractors, and I am putting it into a small number of shops right now, one at a time, done right. If your phone goes to voicemail after 5pm and you know some of those calls are jobs, that is money I can help you stop losing.

Hear it answer a live call or book a 30-minute demo at dispatchbyai.com.

Stop and sit with that. You don’t pardon a man for nothing. You pardon him because you have a reasonable expectation that something is coming. The pardon was the tell. And this week, we found out what it was meant to bury.

On June 18, 2026, on her last day as Director of National Intelligence, Tulsi Gabbard declassified and released hundreds of internal communications from inside the Intelligence Community. The release is titled, plainly, “Fauci Funded Wuhan Lab Research That Sparked COVID.” Read the documents yourself before anyone tells you what to think about them. They are posted on the ODNI site. That is the whole point of a maximum transparency mandate. You don’t need a pundit to interpret them for you.

Here is what the release alleges, and I am going to be careful to call these allegations, because that is what they are until a court or an inspector general says otherwise.

Fauci, as head of NIAID, directed millions in US taxpayer dollars toward dangerous coronavirus research at the Wuhan Institute of Virology. The lab leak that ODNI now describes as the widely viewed source of the pandemic traces back to work he funded. Then, when the origin question became politically radioactive, Fauci did not step back. According to the released emails, he stepped in. He fed hand-picked, NIAID-funded scientists into the Intelligence Community as supposedly neutral experts. Their input shaped the official assessments. Those assessments were then cited publicly as scientific consensus to knock down the lab-leak theory. ODNI calls it a circular reporting loop. I call it laundering your own conclusion through the spy agencies and selling it back to the public as independent analysis.

It gets worse. ODNI says Fauci promoted a paper he helped prompt into existence, a paper they describe as fraudulent, and pushed it to the IC as legitimate evidence. Senior analysts, according to the documents, praised him not as the policymaker and grant-funder he was, but as an unbiased guide to the real coronavirus experts. The fox was not just in the henhouse. The fox was writing the security assessment on henhouse safety.

And the people inside who saw it for what it was got crushed. The release describes whistleblowers who challenged the origin conclusions and were marginalized, threatened with retaliation, and watched their careers stall. A contractor was terminated days after coming forward to ODNI. Managers reportedly reminded analysts that leadership decided who got promoted, and the analysts who pushed the lab-leak hypothesis could read the room. Senior leaders allegedly stripped the anonymity that the whistleblower process is supposed to guarantee by demanding managers or attorneys sit in on the complaint meetings. That is not oversight. That is intimidation with a government badge on it. Gabbard has referred these accounts to the Intelligence Community Inspector General.

Then there is the sworn testimony. On June 3, 2024, Fauci sat in front of the House Select Subcommittee on the Coronavirus Pandemic. He was asked, repeatedly, whether he ever spoke to the FBI, CIA, DIA, or any US intelligence agency about viral research. He dodged, and then he said the words, not to my knowledge about COVID. ODNI says the correspondence they just released directly contradicts that answer. If that holds up, it is not a gray area. Lying to Congress under oath is a felony. He called the cover-up accusations preposterous to the committee’s face.

Now go back to the pardon.

If none of this were true, Fauci would not have needed it. Innocent men do not require a blanket federal shield signed in the dark on the way out the door. The pardon does not prove guilt, and Biden’s own lawyers were careful to write that the pardon is not an acknowledgment that anyone did anything wrong. Fine. But a pardon is the most powerful tool a president has to make a problem disappear, and it was deployed on Fauci before the public had the evidence in hand. The sequence matters. They pardoned him first. The documents came second. You are supposed to never see step two.

Here is the part I want you to feel, not just process.

If these allegations are true, this is not a scandal in the ordinary Washington sense. It is not a bad news cycle. A pandemic killed more than a million Americans and millions more worldwide. It shuttered businesses, including a lot of yours. It put people in the ground alone, with families on the wrong side of a hospital window. If the man who helped fund the research that caused it then used the machinery of American intelligence to bury his own role, lied to Congress about it under oath, and got a preemptive pardon to walk away clean, then we are not talking about misconduct. We are talking about crimes against humanity dressed up as public health, with the cover-up engineered inside the agencies that are supposed to work for us.

I am a veteran. I took an oath to the Constitution, not to a man and not to an agency. The thing that should make every American’s blood run hot here is not partisan. It is structural. The Intelligence Community was turned into a public relations shop for one official’s reputation. Dissent was punished. Whistleblowers, the people the system depends on to catch exactly this kind of rot, were identified and squeezed. And the last move on the board was a pardon that put the whole thing beyond the reach of a courtroom.

So be angry. But be precise about it. The release is a set of allegations and declassified internal documents, not a verdict. Fauci has denied wrongdoing for years and is entitled to that defense. The lab-leak origin remains contested across the agencies, and honest people should say so. None of that lowers the temperature on what the documents describe, because the conduct does not require the origin to be proven beyond doubt. Manipulating intelligence, retaliating against whistleblowers, and lying to Congress are wrong on their own terms, whatever a future inquiry concludes about a wet market versus a lab.

Read the documents. Demand the inspector general finish the job. And remember the order of operations, because it tells you everything. The pardon came first.

Stay sharp.

The United States is punishing a private American company for refusing to do its bidding, and wrapping that punishment in the flag of national security.

The company is Anthropic. In March, a federal judge in San Francisco reviewed the government’s reasons for targeting it and called them what they were: a lie. Judge Rita F. Lin ruled the national security claims “pretextual.” The real motive, she wrote, was “unlawful retaliation.” Nothing in the law, she added, supports “the Orwellian notion that an American company may be branded a potential adversary and saboteur of the United States for expressing disagreement with the government.”

On June 9, Anthropic released Fable 5, the most powerful AI model yet made available to the public. Three days later, on a Friday afternoon, a federal letter arrived. It ordered Anthropic to block every foreign national on Earth from using it, including its own foreign employees. Under export-control rules, that meant one thing: shut it down. Within hours, Fable 5 went dark. Every other Anthropic model stayed online.

The government never specified the security threat in writing. Verbal briefings pointed to a “jailbreak” where users asked the model to review code and fix bugs, the same capability already live in competitors’ models, including OpenAI’s GPT-5.5, which remains fully available and just won Pentagon contracts.

This wasn’t safety enforcement. It was selective punishment.

We Have Done This Before

History is repeating itself with eerie precision.

In 1991, programmer Phil Zimmermann released Pretty Good Privacy (PGP) encryption software. It spread globally. The U.S. government treated him like an arms dealer. Strong encryption was on the Munitions List, alongside cruise missiles. “Exporting” it could mean simply letting a foreigner see the code. Zimmermann endured years of investigation and massive legal bills. No charges stuck.

Critics exposed the absurdity: they printed the RSA algorithm on T-shirts with machine-readable barcodes. Wearing one through an airport was, by the government’s logic, exporting a weapon

.

The controls protected nothing. The technology had already escaped. The rules only burdened Americans who followed them. Those controls eventually collapsed under their own contradictions.

The same pattern is playing out with Anthropic. The capabilities the government claims to fear already exist across the industry. This export order doesn’t contain the risk. It singles out one company that tried to play by the rules.

The Motive Is in the Contracts

Safety was never the issue.

In July 2025, Anthropic became the first frontier lab cleared for classified Pentagon networks under a contract worth up to $200 million. The deal required honoring Anthropic’s policies: no mass surveillance of Americans and no fully autonomous weapons that select and kill targets without human oversight.

Dario Amodei refused to drop those limits. The Pentagon missed its own deadline, and Defense Secretary Pete Hegseth publicly attacked him.

Then, on May 1, the Pentagon awarded major secure contracts to eight other companies: SpaceX, OpenAI, Google, Microsoft, Nvidia, Amazon Web Services, Oracle, and Reflection. All agreed to support any “lawful” military purpose. Anthropic was the only major frontier lab excluded, precisely because it drew a line.

If those limits made a model dangerous, the Pentagon would have rejected the restricted one and embraced the unrestricted. It did the opposite. It rewarded compliance and punished restraint.

This is not how you enforce safety. This is how you pick winners, and punish dissent.

A Federal Judge Already Called It Out

This is not speculation. It is a federal court finding.

Judge Lin pressed the government: If chain-of-command concerns were real, simply stop using the product. No one forced the Pentagon to buy it. Why brand an American company a saboteur instead?

There was no honest answer. On March 26, she issued a strong injunction, calling the actions “classic illegal First Amendment retaliation.” The national security rationale was a cover story.

The administration attacked the ruling anyway. The same pattern, same company, same refusal to remove the same safeguards, has now repeated in June. The court already exposed the pretext once. Decide for yourself if the second round is any more legitimate.

Why This Should Alarm You

This is un-American at its core. A company exercised its right to say no to government demands. The government responded by trying to destroy it.

The damage extends far beyond Anthropic. Every AI lab has now learned the lesson: Write real, binding safety limits into contracts, and you risk being crushed. Make empty promises instead, and you win the deals.

Washington has turned genuine safety into a competitive liability. The next time a company is asked to enable mass surveillance or lethal autonomous weapons, it will remember what happened to the one that refused.

A transparent, fact-based process for blocking truly dangerous deployments can and should exist. This was the opposite: a vague Friday-afternoon order targeting one company while identical capabilities remain available from compliant rivals.

The technology the government claims to fear is still on the open market. The only thing removed was Anthropic.

That was not the free market at work. That was Washington choosing the winner, and punishing the company that dared to say no.

Stay sharp

.

This was not a bug or an outage. The government reached in and turned it off.

THE FACTS

Two models are gone. Claude Fable 5 and Claude Mythos 5. Disabled for every customer at once.

The order came from Commerce Secretary Howard Lutnick, sent straight to Anthropic CEO Dario Amodei. It bars any foreign national from using the models, inside the United States or out. Anthropic got the order Friday evening and pulled both models the same night.

Fable 5 had been public for only a few days. It was the first Mythos-class model Anthropic released to the public. Mythos 5 was never public at all. It ran inside a restricted program called Project Glasswing, open to roughly 50 vetted organizations. Amazon. Apple. Google. Microsoft. CrowdStrike. That model had sped up parts of drug design by about ten times. Mozilla used Mythos-class capability to fix hundreds of software vulnerabilities. All of that is now paused.

Thousands of businesses built on the Fable 5 API lost access with no warning. The API string claude-fable-5 returns an error. Opus 4.8 is the fallback. Every other Anthropic model still runs.

WHAT THE GOVERNMENT ACTUALLY DID

It used export-control authority. Export control covers goods and technology shipped to other countries. Fable 5 was not shipped anywhere. It runs on servers inside the United States.

The order does not restrict where the software goes. It restricts who is allowed to use it, by nationality. By Anthropic’s own account, that includes Anthropic’s foreign-born employees sitting at desks in the United States.

Anthropic said that if this rule were applied across the industry, it would stop every frontier company from releasing new models. Reporting calls this the first hard enforcement of the AI executive order. If the government can switch off a model with a letter, it can switch off any of them.

NO REASON WAS GIVEN

Anthropic says the letter offered no explanation. None. The company’s own guess is that the order traces to a narrow jailbreak, a trick that involves asking the model to read a specific piece of code. It was not a universal jailbreak. The models had gone through thousands of hours of red-teaming by the U.S. government, the U.K. AI Security Institute, and outside teams, and none of them found one.

Want to read the order and judge for yourself? You can’t. Nobody outside Anthropic and the Commerce Department has seen it. Anthropic says it will share more within 24 hours and is trying to restore access.

WHY THIS MATTERS

I build on these tools. I run an SDVOSB and I work in regulated and federal environments. When a vendor kills a tool, that is an outage I have to explain to clients. Friday night a model I was using stopped existing because of an order I am not allowed to read.

This is the whole case for owning your infrastructure. The hardware in my own building does not get a letter from the Commerce Department. A cloud model can be gone before morning.

The lesson is simple. If someone else can turn your tools off, they are not your tools. Own what you can. Have a plan for what you can’t.

I cover this daily on my Substack and YouTube. Follow me everywhere as @vincentactual. vincentactual.com.

The film went to the Navy’s Photographic Interpretation Laboratory at Anacostia. The lab spent more than one thousand man-hours on frame-by-frame analysis. The written conclusion: not aircraft, not balloons, not birds. Unidentified.

Six months later, five physicists watched the same film for part of one morning in a Pentagon conference room and called it seagulls.

That is the Robertson Panel.

The Central Intelligence Agency convened it, not the Air Force. On September 24, 1952, H. Marshall Chadwell, Assistant Director of the CIA’s Office of Scientific Intelligence, signed a memorandum classified Secret to Director of Central Intelligence Walter Bedell Smith. The subject line read Flying Saucers.

Sightings of unexplained objects at great altitudes and traveling at high speeds in the vicinity of major U.S. defense installations are of such nature that they are not attributable to natural phenomena or known types of aerial vehicles.

The CIA’s own Assistant Director for Scientific Intelligence put on paper, to the Director, that the sightings clustered over American defense installations and could not be explained as natural phenomena or known aircraft.

In the next paragraph of the same memorandum, Chadwell defined the security problem.

Public concern with the phenomena indicates that a fair proportion of our population is mentally conditioned to the acceptance of the incredible. In this fact lies the potential for the touching-off of mass hysteria and panic.

Chadwell’s worry was Soviet exploitation. A first strike could be masked behind a flood of UFO reports clogging the air defense channels. He wanted Smith to authorize a program to reduce the reports.

The CIA’s Intelligence Advisory Committee approved on December 4, 1952. The panel met for four days starting January 14, 1953.

Howard Percy Robertson presided.

If you have been reading along, his name has come up before. In May, when I walked through the SHAEF Air Technical Section file at the National Archives, one sentence near the bottom of a 1944 memo from Air Commodore C. M. Grierson read: Mr. Robertson has also been asked to consider the problem.

Around him sat Samuel Goudsmit, Luis Alvarez, Lloyd Berkner, and Thornton Page. Every man held active clearances. Every man had spent the war inside compartmented programs. The CIA pulled the senior bench of America’s classified scientific establishment.

Frederick Durant of CIA scientific intelligence sat as recorder. The unclassified summary he produced is on the public record. The operative language is in Tab A.

The headline conclusion has been quoted every time the Robertson Panel has been written about for the past seventy-three years.

The evidence presented on Unidentified Flying Objects shows no indication that these phenomena constitute a direct physical threat to national security.

It does not say the phenomena are not real. It says they are not a direct threat. The recommendation the panel actually wanted the agencies to act on is in the next paragraph.

We strongly recommend that the national security agencies take immediate steps to strip the Unidentified Flying Objects of the special status they have been given and the aura of mystery they have unfortunately acquired.

Strip the special status. Strip the aura.

Then the panel told them how.

The Panel’s concept of a broad educational program integrating efforts of all concerned agencies was that it should have two major aims: training and “debunking.”

The word debunking, in quotation marks, was written into a formal CIA-approved recommendation in January 1953. The panel put the marks there. They made debunking one of two named aims of an official federal program.

The plan ran through mass media, television, and motion pictures. The report named Walt Disney Inc. as a candidate producer of educational films, a suggestion made at the table by J. Allen Hynek himself.

And then this.

The Panel took cognizance of the existence of such groups as the “Civilian Flying Saucer Investigators” (Los Angeles) and the “Aerial Phenomena Research Organization” (Wisconsin). It was believed that such organizations should be watched because of their potentially great influence on mass thinking if widespread sightings should occur. The apparent irresponsibility and the possible use of such groups for subversive purposes should be kept in mind.

Two American civilian research groups, named on the page, marked for federal surveillance in a CIA-approved recommendation.

I spent eight years in the Navy as an Operations Specialist, four of them aboard USS Peleliu in Tactical Air Control Squadron 11. Internal truth and public messaging are not the same product in any serious military operation. You learn to keep them separate or you do not last on the watch.

The Robertson Panel did not touch the classified record. Chadwell’s defense-installation memo stayed in the CIA’s file. The Newhouse film stayed on the Navy’s books at unidentified. The panel was hired to design the public face. The investigation kept running underneath.

In December 1969, the Air Force closed Project Blue Book and announced the government was finished with UFOs. The FBI’s file on the same subject, 62-HQ-83894, had been open since 1947 and stayed open. In September 2023, an FBI agent at a restricted military range filed a UAP report on Form FD-302. The custody-control language at the bottom of that report was the same language printed at the bottom of FBI Special Agent D. Arthur Byrnes Jr.’s May 1964 Socorro memo.

The Atomic Energy Act of 1946 built the federal mechanism for permanent secrecy. Certain technical facts get classified automatically, beyond the reach of courts and Congress without specific authorization. The Robertson Panel showed that by 1953, the same mechanism was already being applied to UFOs.

AARO is the All-domain Anomaly Resolution Office. Congress created it in the 2022 defense authorization and required it to release UAP case files to the public. AARO released its first one in May 2026. That was the first government UAP report in seventy-three years without the debunking frame the Robertson Panel designed.

Congress did not give AARO the authority to declassify what sits underneath. The Schumer-Rounds UAP Disclosure Amendment would have. It cleared the Senate and was gutted in conference in 2023.

The Durant Report is on the public record. So are the September 1947 Twining memorandum, the May 2026 AARO case file, and the 2023 FD-302 from FBI file 62-HQ-83894. Read them tonight, side by side.

Tab A is still operative. The open question is what has been filed under it for seventy-three years, and who has been cleared to read it.

Stay sharp.

This series exists because readers fund it. Become a paid subscriber to keep the documents coming.

The agent, listed in the file only as USPER5, spotted what looked like a car ahead. Their partner, USPER6, was with them. The “car” had two lights, one red, one white, at normal headlight height.

They treated it as a routine trespass and gave chase.

They got within a few hundred feet. Then the object did something strange.

It did not turn. It slid sideways off the road at 15 to 20 miles per hour. It moved with “zero resistance,” according to one witness. It stopped about 100 meters away and turned its lights off.

USPER6 put on night vision goggles. The object looked like a thin horizontal line, about four feet wide. The other agent saw a dark kite shape with rounded sides. Later, both described it as triangular.

The key point is this. The agents acted like normal patrol officers. They saw something where it shouldn’t be, so they chased it and used their tools. The strangeness came from what they found.

Thirty minutes later, another report came in. The same two agents returned with a third colleague, USPER7.

They saw the kite-shaped object again. This time it was hovering six meters off the ground. The lighting matched the first sighting. USPER7 saw nothing.

Through night vision, USPER5 could see distant stars faintly visible through the object. It was partially transparent

Then it disappeared.

USPER5 grabbed a powerful spotlight and swept the beam across the desert. About fifty yards out, the light beam suddenly stopped, as if it hit a wall. When he moved the beam away and brought it back, the light passed through normally.

They never saw the object again.

That description comes from a slide the Department of War approved for public release.

The File

This report comes from the All-domain Anomaly Resolution Office, known as AARO. Congress created it in 2022. It is now part of the Department of War and is required to release public case files on a regular schedule.

The director who signed this file is Dr. Jon T. Kosloski. He is a quantum-optics physicist. He came from the NSA, where he led signal-detection work in the Research Directorate and came up through the Cryptanalysis Development Program. His doctorate is in electrical engineering from Johns Hopkins. The May 8 file carries his name.

The document is short, just four pages. It covers four incidents over three days. Seven federal agents are mentioned (USPER1 through USPER7). Their real names and agencies are hidden. The exact locations are hidden too.

What is not hidden is AARO’s own scientific measurements.

This is the first time since 1969 that the U.S. government has released a case file in this clear, structured format.

What the File Shows

The kite-shaped object gets the most attention. But the strongest cases are the first two.

The Orange Orbs

On two straight evenings at dusk, three separate teams of agents saw the same thing from different locations.

An orange orb appeared in the sky. It released several smaller red orbs that flew away horizontally. Then the orange orb vanished. This pattern happened at least five times.

Six trained agents. Independent views. That is strong evidence.

The Large Fiery Orb

In another sighting, USPER5 and USPER6 saw a glowing orange object hovering near a rock pinnacle. They thought it was 500 to 600 meters away and about the size of a small helicopter cockpit.

The file calls it a “Large, Fiery Orb.” One agent said it looked like the Eye of Sauron.

Here’s the important part.

AARO later went back to the site and took follow-up measurements using the rock pinnacle as a fixed reference point. Their instruments and analysis showed the object was actually about 1,050 meters away, nearly twice as far as the agents had estimated and 12 to 18 meters wide (roughly the size of a small house or large truck).

The witnesses said “small and close.” AARO’s refined data said “large and far.” The file prints both estimates side-by-side on purpose.

Read what that does. AARO didn’t explain the object away or dismiss it as a mistake. Instead, they made the sighting more specific and concrete. The corrected picture is of a 12-to-18-meter glowing orb hovering motionless right next to a rock pinnacle at roughly one kilometer’s distance, until it suddenly wasn’t.

That’s the federal government’s own measurement tools and site analysis adding weight to the strangeness, not lifting it.

The Real Change

The incidents matter. But the bigger story is how AARO handled them.

They investigated. They measured. They published.

This is the same careful work the FBI did in 1964. The difference is in how it was released.

Byrnes’s 1964 memo landed in FBI headquarters file 62-HQ-83894. At the bottom of every page was a custody stamp declaring the file was FBI property, loaned only to the receiving agency and not to be shared further.

The May 8 AARO file carries no such stamp. It belongs to the Department of War — and it is now directly in the hands of the public.

That is not a minor paperwork detail. For decades, reports like these carried restrictive federal custody language. They lived inside permanent files the public was never meant to see. The May 8 release marks the first time in your lifetime that the kind of detailed report trained federal agents file from inside a closed federal zone has reached ordinary citizens without those barriers.

The process stayed the same. The openness is new.

Where It Happened

The file only says “Western U.S.” Specific details are redacted.

What stands out is the pattern. Federal agents in teams. Inside restricted zones. Using night vision and spotlights. Responding to objects on closed roads.

This matches the heavy security areas in the American Southwest, especially around Sandia, Kirtland, and Los Alamos.

Why This Matters

AARO is real progress. The May 8 file proves the office can produce clear, credible public reports.

But AARO is not the full picture. It sits on top of a much older, more secretive system, the one this series has been tracing back to the Atomic Energy Act of 1946. That deeper system still controls most of the information. None of its statutory protections were undone when AARO was created.

So this is where you sit.

You are looking at one slice of what your government knows about objects in restricted airspace. The office that published this slice answers to Congress. You have a vote that reaches it. The architecture under it does not answer to anyone you can vote for, and has not for nearly eighty years.

The May 8 file is what they were allowed to show you. The next question is what they are not showing you.

Stay sharp.

Byrnes finished his work, got in his car, and drove out to the site. He was not the only government man who did. By the time he reached the arroyo, New Mexico State Police Sergeant M. S. Chavez was there, along with Socorro County Undersheriff Jim Luckie and State Police Officer Ted Jordan. Within hours an Army captain, Richard Holder, drove down from the Stallion Site at White Sands Missile Range. Within days the Air Force sent its scientific consultant, the Northwestern astronomer J. Allen Hynek. Four institutions converged on one patch of desert over a single sighting.

They did not converge because of a light in the sky. Lights in the sky do not pull the FBI, the Army, the state police, and an Air Force scientist to one arroyo. They converged because there was something on the ground, and because by 1964 each of those institutions had a standing channel for exactly this kind of report and a reflex to feed it.

This piece is about one of those channels. Not the craft. The file the Bureau opened on it, and the fact that the file is still open.

Hoover Wanted the Discs

In July 1947, weeks after the Roswell summer, a memo crossed J. Edgar Hoover’s desk. It asked whether the Bureau would help the Army Air Forces run down the flood of flying disc reports coming in from across the country. Hoover wrote his answer in the margin in his own hand, and the annotation has been quoted ever since. He would assist. But, he wrote, “we must insist upon full access to discs recovered.” Then a grievance, in the same hand: in one case the Army “grabbed it and would not let us have it for cursory examination.”

Read what that note assumes. The Director of the FBI, in the summer of 1947, is not asking whether the discs are real. He is treating recovered discs as physical objects, demanding the right to put his own people on them, and resenting another service for reaching one first. That is not a man humoring a craze. That is a man staking a claim.

The claim became a permanent file. The Bureau opened headquarters file 62-HQ-83894, and field offices fed it for decades. By the time a Socorro patrolman named Lonnie Zamora keyed his radio in 1964, that file had been accumulating flying disc material for seventeen years. Socorro did not begin the Bureau’s involvement. It was one serial dropped into a file that was already old.

Standing

I spent eight years on active duty in the Navy as an Operations Specialist, four of them aboard USS Peleliu in TACRON 11, running the Fleet Air Defense Identification Zone in the Pacific. The work is contact resolution. Something appears in the airspace you are responsible for, and you decide what it is, friendly, hostile, or unknown, and what you do about it.

The radar was never the hard part. The hard part was the witness. A contact is only as good as the person reporting it and the evidence that backs the account. Half the job is judging reliability under pressure, deciding whether what a watchstander swears he saw squares with what the scope and the physical world will support, and knowing when an unknown has crossed the line from anomaly to something you put in writing and send up the chain.

That judgment is exactly what Byrnes exercised at Socorro. He sized up the witness, he measured the evidence on the ground, and he committed both to the federal record. I have read a great many reports built that way, and written my share. This one is clean. The man who built it knew the difference between a witness you bury and a witness you certify, and he chose to certify.

What Byrnes Put in Writing

The memo is dated May 8, 1964, on the letterhead of the United States Department of Justice, Federal Bureau of Investigation, Albuquerque, New Mexico. The subject line is three flat lines of capitals: UNIDENTIFIED FLYING OBJECT, SOCORRO, NEW MEXICO, APRIL 24, 1964. It is written in the register of a man who documents scenes for a living, and the restraint is the evidence.

Byrnes measured the ground where the object had stood. He recorded four indentations in the rough earth, each roughly sixteen by six inches, rectangular, about two inches deep, each pressed into the soil at an angle from a center line, as though something had set down on canted supports and shoved the dirt to the far side. Inside the rough rectangle those four marks framed, three burned clumps of grass. Other clumps in the same area, untouched. Three smaller circular marks, smooth, about four inches across, sunk perhaps an eighth of an inch into the sand “as if a jar lid had gently been pushed into the sand.” No dwellings in sight. No other objects. No other person in the area that night.

And then the sentence that carries more weight than anything else in the file. Byrnes had known Zamora personally for about five years. He put the assessment in the memo, on the record: the officer “is well regarded as a sober, industrious, and conscientious officer and not given to fantasy,” and on the night in question was “perfectly sober and somewhat agitated over his experience.”

A federal agent does not write “not given to fantasy” about a man he intends to discredit. He writes it about a man he is vouching for. Byrnes was spending his own credibility, in writing, in a document bound for a permanent Bureau file, to certify that the witness was reliable and the marks in the dirt were real.

At the bottom of the page sits a block of boilerplate that is easy to skip and worth reading slowly. The document, it says, “contains neither recommendations nor conclusions of the FBI. It is the property of the FBI and is loaned to your agency; it and its contents are not to be distributed outside your agency.” Hold onto that paragraph. It comes back.

What Zamora Reported

The account in the file is Zamora’s own, taken the night of the 24th and into the morning of the 25th. Strip away the mythology that grew around the case in the decades after and read only what the patrolman told the men standing in the arroyo with him that night.

He had been chasing a speeding black Chevrolet south out of town when he heard a roar and saw a flame in the sky to the southwest, bluish and orange, narrow, funnel-shaped, descending slowly. His first thought was not a spacecraft. It was a dynamite shack he knew sat out that way and the chance it had blown. He broke off the chase and fought his patrol car up a steep gravel hill, wheels skidding, three tries before it took the grade.

Over the top he saw a shiny object 150 to 200 yards off the road. At first he read it as a car flipped onto its roof, kids who had rolled it, and beside it he saw what looked like two people in white coveralls. One of them turned, looked straight at his patrol car, and seemed to start, as if surprised. He radioed it in as a possible accident and said he was getting out to check.

It was not a car. As he got out he heard two or three loud thumps, like heavy doors slamming a second apart. Then the roar, climbing from low frequency to high, very loud at that range, and Zamora knew jet engines and stated flatly that this was not one. A flame came from underneath the object, light blue going to orange, and the thing rose straight up. Smooth. No windows, no doors he could make out. Aluminum-white against the mesa. And on its side, in red, an insignia he estimated at about two and a half feet high and two feet wide, set in the middle of the body.

He thought it might explode. He ran, struck his leg on the patrol car’s rear fender, lost his glasses in the dirt and left them. He put the car between himself and the object. When he looked back it had risen to about the level of the car, twenty to twenty-five feet, then it cleared the terrain and went off to the southwest. The roar cut out and left a high-to-low whine behind it, and then nothing.

A named officer. On duty. Sober. Vouched for by an FBI agent who had known him five years. Describing a landed object with markings and two occupants, with physical traces in the ground that the agent then measured with his own hands.

Three Agencies, One Arroyo

This is where Socorro stops being a UFO story and becomes a structure story.

Go back to who was standing in that arroyo. Holder, the Army captain, was the up-range commander at the Stallion Site of White Sands Missile Range, and he took his own report that night alongside the FBI’s. Hynek came down for the Air Force through Project Blue Book, walked the site, interviewed Zamora, and came away regarding him as entirely credible. Blue Book worked the case for months and never explained it. It remains in the Air Force’s own records as one of the relatively small number of cases the program formally logged as unidentified.

Now count the channels. The New Mexico State Police on the ground first, through Chavez and Jordan. The FBI documenting and certifying, through Byrnes. The Army taking a parallel report off a strategic missile range, through Holder. The Air Force running a scientific investigation through a standing national program, through Hynek. Four institutions, one patch of desert, one afternoon, and not one of them had to invent a procedure on the spot. Each already had a channel and a form and a file to put this in.

That is the tell. The procedures already existed. In 1947 the Air Force had been directed to collect, collate, evaluate, and distribute reports of the phenomenon, and to route them to interested government agencies and contractors. By 1948 that pipeline had pulled in the Army and the FBI. Socorro in 1964 is what that architecture looked like after sixteen years of operation. Not a scramble. A reflex. Four agencies converging on a landing site because converging on landing sites was, by then, simply what the machinery did.

The Program That Closed, and the One That Didn’t

Five years after Socorro, the public-facing half of that machinery shut down.

In December 1969, on the strength of the University of Colorado’s Condon Report, the Air Force announced the termination of Project Blue Book. The official line was clean and quotable. No UFO reported, investigated, and evaluated by the Air Force had ever given any indication of threat to national security. There was no evidence the sightings represented technology beyond present knowledge, and none that anything seen was extraterrestrial. The government, the announcement made clear, was getting out of the UFO business.

That was the posture. Blue Book was the program the public could see, so closing Blue Book was the act the public could see. The headlines in January 1970 all said the same thing in different fonts. The Air Force is done with flying saucers.

The FBI’s file did not close. There was no announcement, because you cannot hold a press conference shutting down a channel you never advertised. Blue Book had always been the storefront. The Bureau’s 62-HQ-83894 was something else, a permanent records function that predated Blue Book and answered to a different chain entirely. When the storefront closed, the back office kept its lights on.

Set this beside what the 1947 record already showed. In December of that year the Air Force ran two documents through the same headquarters in the same week. One was the internal record, an active intelligence matter, qualified observers, a real phenomenon, sent to the Chief of Staff. The other was the external posture, a civilian’s photographs dismissed as “film defects” with an order that no further investigation be made. Same office, same week, two different faces. The 1969 Blue Book closure is that same maneuver run at the scale of an entire program. The public face announced an ending. The internal capability moved out of view and kept running.

Fifty-Nine Years Later

Move the file forward to September 2023.

A team of contractors was running LiDAR tests on a restricted military range. The lead, a woman with fifteen years on that range, had personally closed the airspace for the tests that morning. At a gate that had never given trouble, the access fob failed three times and opened on the fourth. Just past it, driving to the test site, she looked up and saw a metallic object to the southwest, the length of two or three Blackhawk helicopters nose to tail, with an intense diamond-white light on one end that she compared to looking into the sun. It was nearly hovering, drifting slowly east to west, completely silent. After five to ten seconds it was simply gone from a clear sky. A contractor in the second vehicle saw it too.

The reason any of that exists in writing is that the FBI took it. Agents interviewed the witnesses in October 2023 and wrote them up on Form FD-302, the Bureau’s standard report of interview. And at the bottom of those 2023 pages sits a paragraph worth reading word for word against the one Byrnes signed in 1964. The document, it says, “contains neither recommendations nor conclusions of the FBI. It is the property of the FBI and is loaned to your agency; it and its contents are not to be distributed outside your agency.”

It is the same sentence. Not a paraphrase, not a descendant. The identical custody language, on an FBI report of a credible witness and an object that should not have been there, fifty-nine years after Byrnes drove out to the arroyo. The Bureau was documenting the same kind of report, asserting the same ownership over it, and loaning it out under the same restriction, in 2023 as in 1964.

That is the continuity made literal. Not a theme I am imposing on two cases. The same custody stamp, the same act of an agent committing a witness’s account to a federal file the Bureau controls. The channel Hoover opened in 1947 was still open in 2023, fifty-four years after the Air Force told the country the government had stopped looking.

Why It Matters

The disclosure conversation files the FBI’s role under trivia. A famous Hoover note, a few yellowed memos, a cabinet nobody opens. The record describes something far more durable than trivia.

The function the Air Force was ordered to perform in 1947, collect the reports, evaluate them, route them, and keep the serious internal record separate from the public posture, did not stay in the Air Force and it did not stop when the Air Force said it had. By 1948 it had branched into the Army and the FBI. By 1964 the FBI was running it as independent field practice, with its own permanent file, its own agents at landing sites, its own custody rules, while four agencies converged on a single arroyo as routine. In 1969 the visible program was closed with a press release. And in 2023 the Bureau was still doing the work, on the same form, under the same custody stamp, for an object over a restricted military range.

That is a single institutional capability operating for seventy-six years, across at least three federal agencies, surviving the public shutdown of the one program citizens were allowed to watch, feeding a records system the public was never structured to reach. Capabilities that run that long and that quietly do not run on inertia. They run on a legal framework that instructs the institution to keep collecting, keep filing, and keep most of it out of public hands. That framework is the one this series has traced back to the Atomic Energy Act of 1946 and the contractor special access program structure built on top of it.

The constitutional question is not whether Lonnie Zamora saw a craft, or whether a contractor saw one in 2023. It is whether a reporting and classification capability that has run without interruption since 1947, across agencies, through the announced death of its own public face, into a system Congress cannot fully reach, has outgrown the Article I oversight that is supposed to govern it.

An FBI agent measured the marks in the dirt at Socorro in 1964 and signed his name to the man who saw them. The Bureau filled out the same form, under the same stamp, for the same kind of report in 2023. The Air Force told the public it was finished in 1969 and the file stayed open the whole time. The instrument has not changed in fifty-nine years. The open question is who, in all that time, has been cleared to read what the Bureau has been keeping.

Stay sharp.

Colonel H.M. McCoy ran intelligence at the Air Materiel Command. He reported to Lieutenant General Twining. In December 1947 he had bundled three months of accumulated flying disc material and sent it up the chain to the Chief of Staff of the Air Force. Eleven weeks later he came back with an operational request. Put dedicated interceptors on continuous alert inside the country and use them to chase the phenomenon.

On March 9, 1948, Washington answered. The memo came from Lieutenant Colonel George D. Garrett at Air Force Headquarters. Subject line, “Flying Discs.” The opening reference is to “Colonel McCoy’s informal proposal that certain fighter aircraft be maintained on a continuous alert status.”

The answer was no. Read the reasons, because they are not what you would expect.

The aircraft and personnel would cost too much against the results. Interception was not possible without complete radar coverage. The Air Force could not provide that coverage. Even if the fighters launched, the odds of intercepting the objects were doubtful.

That is not a man saying the discs are imaginary. That is a man saying we cannot afford to chase something we can’t yet track. Two officers, one document. McCoy ready to commit fighter assets. The Air Staff blocking it on infrastructure, not on substance. Both of them treating a phenomenon the institution had formally called real five and a half months earlier as a live air defense problem.

This is what it looks like when an intelligence program goes operational.

My Standing

I spent eight years on active duty in the Navy as an Operations Specialist. Four of them aboard USS Peleliu in TACRON 11, running the Fleet Air Defense Identification Zone in the Pacific. The watch I stood is the same watch the 1947 Air Materiel Command was standing. Identify what is in the air. Call it friendly, hostile, or unknown. If it is unknown, you escalate. By the spring of 1948 the institution that had stood that watch was running a global reporting pipeline. The 1948 packet is what the pipeline carried.

What Happened in the First Five Months

The December 30, 1947 directive ordered the Air Materiel Command to set up the project that became Project Sign. It told AMC to collect, collate, evaluate, and distribute everything on the phenomenon to “interested government agencies and Contractors.” That is the document everyone cites.

Almost nobody reads what came after it. Twenty-eight pages, thin and dull, the operational receipts of the program the directive created. Watch how fast it stood up.

By February 12, Project Sign had a file number. USAF Headquarters filed a reference memo to AMC indexed under decimal 000.9, “Flying Discs,” summary line “Project ‘SIGN’.” The next day a parallel card went in under “Phenomena.” Two permanent entries in the Air Force’s central filing system inside one week.

A name. A number. A standing program with a paper trail.

Then it went worldwide. On February 25, Garrett restated the policy to every Air Force installation on earth. The two lines that matter:

AMC “has been designated the Air Force agency to collect, collate, evaluate, and distribute to interested government agencies and contractors all information concerning sightings.” And AMC “has been authorized direct contact with all Air Force installations whenever required to develop additional information on the flying discs.”

Strip the bureaucratic skin off that. Wright Field is the clearinghouse. The contractor distribution loop is standard procedure. Every base, inside the country and out, is ordered to report up the chain. And AMC can skip normal channels and call any installation directly to work a case. Two months after the founding directive, the pipe was open at every base in the world.

Then the Reports Came In

What follows in the packet is exactly what a live worldwide reporting system produces. Base intelligence officers filing incidents, routed up through the numbered air forces, into Headquarters, out to Wright Field.

Brooks Field in San Antonio. Hamilton Field in California. Tyndall in Florida, routed through the Air University in Alabama. Harrisburg in Pennsylvania. A foreign-source report from the air attache in Canada. Texas, California, Florida, Alabama, Pennsylvania, Canada. The pipe ran every direction at once.

One report carries a sting. It came from the Fourth Air Force shop at Hamilton Field, the same office that in December 1947 had forwarded a civilian’s photographs with a note calling the source “very reliable.” Washington had killed that case as “defects in the film, paper, or camera” and ordered no further investigation. The intelligence officer was Lieutenant Colonel Donald L. Springer.

On March 11, Springer forwarded a fresh one. Bakersfield, two incidents days apart, multiple witnesses including a Pacific General Electric employee and a Sheriff’s sergeant. Objects falling from unknown sources, smoke and debris trails. Search parties, aircraft, and rescue units went out. Nobody found anything.

Sit with that. The same office told three months earlier to stop investigating was still routing live multi-witness reports up the chain. The public dismissal posture had not stopped the work. It had only controlled which cases the public would ever hear about.

Then Everyone Wanted In

By late March the architecture started crossing service lines. On March 25, the Director of Intelligence for the Army General Staff asked to be put on distribution for every flying disc communication touching Army-subordinate commands.

The Army had been a separate service for barely six months at that point. Asking for routine intelligence copies on a subject another service was running is not curiosity. It is one institution telling another that it considers this its business too.

Then the Bureau showed up. On May 8, near Mansfield, Ohio, four New York Central Railroad employees watched a phosphorescent object cross the sky at “a great amount of speed,” trailing light. Named men. A car inspector, a yard clerk, a railroad patrolman. Their statements did not enter through a base intelligence office.

They came in through the FBI. Special Agent D.K. Brown out of Cleveland took the witnesses, and the report worked its way through the Eleventh Air Force to USAF to Wright Field. The Bureau would not open its own permanent UFO file for another three years. It was already feeding the Air Force pipe in 1948.

What the Twenty-Eight Pages Actually Prove

There is no smoking gun in this file. Filing cards, cover memos, a rejected fighter proposal, routine reports from five states and a foreign desk. That is the point. Working programs do not produce smoking guns. They produce paperwork.

And the paperwork is damning on its own terms. File number inside sixty days. Worldwide reporting policy inside sixty days. Fighter alert proposed inside ninety. Army on distribution inside four months. FBI feeding the pipe inside five. All of it before the 1948 fiscal year was halfway done. Every piece the Atomic Energy Act framework would later formalize and make permanent was already running. The contractor loop. The cross-agency routing. The global reporting requirement. The clearinghouse at Wright Field.

The disclosure debate gets argued as if the government spent eighty years failing to figure out how to handle this. It did not. The handling was settled in the first five months and carried forward ever since.

So the constitutional question is not whether the Air Force took the matter seriously. The record settles that. The question is whether the architecture they built that fast, later folded into the Atomic Energy Act classification system and extended through the contractor special access program world, has grown an oversight gap that Congress can no longer reach.

McCoy wanted fighters on continuous alert seventy-seven years ago. He did not get them, for radar reasons. The architecture he wanted those fighters to feed is still running.

The radar has gotten better since.

Stay sharp.

That’s the first thing my BM3 said to me after I almost drowned. Get back to work. You aren’t taking lunch yet.

I was 22. USS Hue City. CG-66. Ticonderoga-class cruiser out of Mayport, Florida. I’d enlisted the year before with a guaranteed A-school for Operations Specialist. The rate that runs the radar picture in CIC. That was the deal I signed at MEPS.

Here’s what they don’t tell you at MEPS. When you sign for a rate, your school seat doesn’t always open up. You ship to the fleet first. You sit. They give you a job while you wait. That job is deck division. I waited nine months. Nine months chipping paint, sanding decks, painting bulkheads. The Navy calls it undesignated. What it actually means is you are disposable labor with a class date in the future that nobody around you cares about.

Deck division on a Navy ship is its own ecosystem. The Boatswain’s Mates run it. The BMs. They are proud of their rate. They have to be. Because nobody else on the ship respects it. The OS’s looked down on deck. The IT’s looked down on deck. The HM’s looked down on deck. Even the cooks looked down on deck. The BMs ran the division like they were defending a fortress nobody else respected. Junior sailors paid the price for that the whole way down.

One afternoon, my BM2 sent me over the side to paint the hull. The tool for that is called a boatswain’s chair. Here is what it actually is. A two-by-four with two holes drilled in it. A rope runs through the holes up to a single point above your head. You sit on the two-by-four. You hang off the side of the ship. You paint.

I didn’t tie the knots. Someone else did. That is how it worked. The chair gets rigged, you climb in, you go over the side. I trusted the rigging because the system told me to trust the rigging. That was the first mistake.

The knot failed. I dropped. Pier-side, not underway. That is the only reason you are reading this. I hit the water between the hull and the pier. The drop was roughly 25 feet. The water was filthy. The current was nothing because we were tied up. Everything that could have gone right in a fall like that did go right. Except for the next thing.

My safety vest did not inflate. The Navy issues you a flotation device that is supposed to deploy automatically when it hits the water. Mine did nothing. Two pieces of safety equipment in one afternoon. Both of them failed. The chair I was sitting in. The vest I was wearing. Either one alone could have killed me. Both at once on the open ocean and I am not here.

I am a good swimmer. Have been my whole life. I swam to the pier and climbed up the ladder. Nobody from the ship had moved yet. By the time anyone realized what happened, I was already on the pier dripping, walking back toward the brow.

I should have gone to medical. Anyone who has fallen 25 feet should go to medical. But medical was frowned on in deck. If you went to medical, you were soft. You were trying to get out of work. You were a problem. So I went back to my berthing, took a shower, put on clean clothes. I told myself I was fine. The deck taught you to eat it. That is what you did.

By the time I was cleaned up, it was lunch. The mess decks were open. I got in line with everybody else. Tray in my hand. My BM3 saw me from across the space. She yelled my name. The whole mess decks heard her. She said get back to work. You aren’t taking lunch yet.

She did not know I had just been in the water. She did not need to know. Yelling was the resting state.

This is the line that explains deck division. Your equipment can fail twice in one afternoon and your NCO will still yell at you for taking lunch. That is the rate. That is the culture. That is what it actually was.

I figured out the escape hatch by month three. Mess crank. You volunteer for 90 days washing dishes for 300 sailors. Grease trap. Trays. Pots. Three meals a day. It sounds miserable. It is miserable. But it is not deck. Nobody yells at the mess crank the way they yell at deck. The job is hot and gross and physical, but you finish your shift and you are done. I volunteered the first chance I got.

Nine months in, my class date hit. I got my orders for OS A-school in Dam Neck, Virginia. I packed a sea bag. I checked off the ship. I never saw that BM3 again. The Navy I went to next was a different Navy. CIC instead of chipping paint. Radar consoles instead of paint rollers. A rate that the rest of the ship actually respected. Same uniform. Different world.

THE LESSON

Here is what I want you to take from this. In deck division, your equipment can fail twice in one afternoon and your NCO will still yell at you for taking lunch. That is the rate. That is the culture. That is what 20-year veterans of the BM community built and what every undesignated kid coming through the fleet has to survive.

I survived it because pier-side geometry. Because I could swim. Because I figured out the escape hatch. Most of those kids didn’t drown. They just absorbed the contempt and carried it the rest of their careers.

Every system has its deck division. Every organization has a place where new people get warehoused while leadership figures out what to do with them. Every culture has a layer where the contempt rolls downhill and the people at the bottom just eat it.

The job, when you find yourself there, is to identify the escape hatch and take it. The hatch is rarely glamorous. Mine was washing dishes for 300 sailors. But it gets you out. And out is where the next thing happens.

I lived it. I got out. I am telling it now.

If this is the kind of story you want more of, I put out daily analysis on Substack and breakdowns on YouTube. The latest video is the full story above. Subscribe if you want the rest.

Follow me everywhere as @vincentactual.

Most of that growth came from one tactic almost nobody in the natsec, UAP, or macro Substack cluster is using deliberately: replying to other people’s Notes with substance.

Not spam. Not link drops. Substance. Real voice. Real engagement. The kind of reply that makes a stranger curious enough to click through to my profile.

Then I built a Claude-powered skill that runs the workflow at scale.

This post is the case study. The numbers, the method, the mistakes, and the tool.

What Substack Notes Actually Is

Substack Notes is a Twitter-style feed inside Substack. Most writers treat it like an afterthought. Random thoughts, occasional screenshots of their latest post, low-effort engagement bait.

What Notes actually is: the highest-leverage discovery surface on the platform.

When you reply to someone else’s Note with a link to your article, Substack auto-renders a link card showing your article’s title and lead image. That card converts better than a bare URL would on any other platform. Substack designed it that way because they want users to discover content without leaving the app.

The math: popular Notes have 50 to 300 visible comments. Each comment is a free distribution channel into someone else’s audience. If your reply is substantive and your article is good, a meaningful percentage of those readers click through and a smaller percentage subscribe.

Compound that across three or four replies a day on the right Notes, and the curve accelerates.

The Workflow

I ran one manual round, then built the skill and iterated through three more rounds in a single day. Here’s what works.

Open the Substack Notes feed. Look for Notes that explicitly invite outreach. “Share what you’re writing.” “Drop your link if you’re new.” “Who do you write for?” These are creators inviting engagement. They are the highest-fit targets.

Skip the wrong targets. Archived posts more than 60 days old. Off-topic feeds. Notes you’ve already replied to. Notes with no real engagement.

Write replies that answer the actual prompt first, then naturally segue into your work. Sentence one: directly answer their question. Sentence two: one line of substance about what you write. Sentence three: the link. The link card does the conversion work, not your prose.

Vary the phrasing across replies. Substack’s spam classifier sees identical text patterns and suppresses them. Same article, different wording every time.

Pace. Maximum four replies per session. Two-hour minimum between sessions. Never two replies to the same author within a week. These rules are not arbitrary. They are the threshold above which the platform flags you.

Verify. After posting, check your own Likes and Replies tab to confirm the reply landed and the link card rendered. Substack sometimes silently drops replies due to rate limiting.

That’s the whole methodology.

The Numbers

Two weeks ago: 45 followers and 96 subscribers (a chunk imported from prior platforms).

Today: 102 followers, 100 subscribers, growth curve up and to the right with the slope steepening in the last five days.

Open rate on my recent posts: 35.88%. Industry average is 20 to 25%. My audience reads what I write.

Substack itself is now my fourth-largest traffic source. 27 unique visitors in the last 30 days came directly from Notes engagement. Substack network drove 8 of my 25 new subscribers in that window, with 5 of those 8 attributed specifically to Notes

.

The breakout post: “The Disclosure Timeline Is Wrong.” 132 views and five free subs from a single piece. The Notes traffic doesn’t always convert on the article I link. It converts on whatever’s newest. The link card brings them in. The body of work keeps them.

The Tool

I built a Claude-powered skill that runs this workflow.

It reads a profile file (my voice rules, featured article URL, audience description), opens Substack Notes, scores the visible Notes for fit, drafts replies in my voice, posts them via browser automation, verifies they landed, and reports the results.

The methodology is mine. Claude is the engine that runs it at scale. I wrote the workflow logic, the voice rules, the target scoring rubric, and the guardrails. The skill just executes the spec I designed.

Roughly an afternoon to build. Three files. A workflow file with the logic, a profile file with my voice and current featured article, and an install file so I could deploy it.

The skill enforces the pacing rules. Max four per session. Vary phrasing. Skip bad fits. If Substack throws a CAPTCHA, stop. This is not optional. It is the difference between organic growth and getting flagged.

I ran the skill four times today. Twelve replies posted. Comment counters incremented on every target. Subscriber count moved from 96 to 100 visible inside one session, with the dashboard lagging behind real-time reality

.

What I Got Wrong

First round, I targeted four Notes from a feed snapshot. By the time I started replying, the feed had rotated and three of my targets were no longer at the top of anyone’s view. The replies still posted but visibility was cut. Lesson: always work from a fresh screenshot, never from yesterday’s data.

Second round, my first version of the skill used hardcoded click coordinates for the reply textarea. Substack’s modal renders at different vertical positions depending on which Note you clicked. Sometimes Substack also overlays a Drafts dropdown showing previous abandoned attempts. The hardcoded coordinates landed on the wrong element and one reply ended up in the search bar instead of the textarea. Lesson: take a fresh screenshot before every textarea interaction. Never reuse coordinates across modals.

Third round, I sent a DM follow-up to a retired Navy Captain and the newline characters in my message triggered Substack’s send-on-Enter behavior. The message went out three lines too early with formatting mashed together. Substance was intact, polish was off. Lesson: in chat interfaces, never include literal newlines in your message text unless you’ve confirmed how the platform handles them.

I’m leaving these in because they’re the actual story. Building this kind of tool isn’t a clean linear process. It’s iterate, observe, fix, iterate again.

Why This Isn’t Spam

The discipline matters here. Anti-spam guardrails are not optional. They are the difference between organic growth and getting suppressed.

Max four replies per session. Three is the sweet spot.

Two hour minimum gap between sessions.

Never two replies to the same author within seven days.

Vary the wording on every reply, even when pointing to the same article.

Skip targets that don’t genuinely fit, even when the slot is open.

If Substack throws a CAPTCHA or unusual-activity warning, stop immediately.

The point is not to flood the platform. The point is to make the reply itself good enough that strangers click through. The link is the conversion mechanism. The reply quality is what earns the click.

If you can’t make the reply genuinely useful to the original poster’s audience, don’t post it. A skipped slot is better than a forced one.

I designed the workflow against Substack’s terms and operate within them. The replies are mine. The voice is mine. The substance is mine. The skill is a productivity layer over a manual process I ran first to verify it worked. That distinction matters and it’s the line every operator using AI tooling needs to hold.

What’s Next

I’m cleaning up the skill so other Substack writers can use it. Profile-driven configuration means any writer can plug in their bio, voice rules, and featured article URL and run the same workflow on their own account. Tiered options from self-serve digital download to done-for-you setup to ongoing retainer.

I’m also leaning into the substance advantage. My competition in this space is mostly speculation and news aggregation. I do legal-document analysis, FOIA-adjacent work, and primary-source writing. The substance compounds when the distribution is solved.

The plan from here: keep the Notes engagement running daily. Push to 500 free subscribers. Then layer on podcast appearances, original FOIA work, and competitive-challenge posts that take direct positions against weaker arguments in my niche.

If you’re writing on Substack and you’ve been ignoring Notes as a growth channel, this is your sign.

If you want to know when the skill ships, subscribe. The launch goes here first.

That sentence is doing a lot of work.

I spent four years tracking aerial contacts off the Pacific coast as a FADIZ operator with TACRON 11 on USS Peleliu. The gap between a briefed capability and the operational reality is usually where the actual story lives. Orbital compute is sitting in that gap right now.

The press is treating space-based data centers as a near-term play. The physics says otherwise. Three constraints stack up against any frontier AI workload running in orbit. A fourth one kills training outright. To see why, you have to understand one distinction first.

TRAINING VS INFERENCE: THE DISTINCTION THAT MATTERS

Two different things happen with an AI model. They get conflated constantly. Sorting them out is the whole point of this piece.

Training is how you build the model. You take a vast dataset, run it through a neural network, and spend months adjusting billions of parameters until the model produces the outputs you want. Tens of thousands of GPUs running in tight synchronization for months at a time. Hundreds of millions of dollars per training run. Bandwidth-hungry. It happens once.

Inference is how you use the model after it is built. Each inference call is independent. It runs on a single GPU or a small cluster. The GPUs do not need to talk to each other constantly. A single inference call costs fractions of a cent.

The physics in orbit penalizes everything training needs and tolerates what inference needs. Frontier training is what every press release implies when it talks about putting AI in space. Inference is what the physics actually permits.

THE THERMAL PROBLEM

In a terrestrial data center you blow air over the chips, run water through cold plates, or dunk the whole rack in dielectric fluid. In vacuum you have none of that. You can only radiate heat. The math is Stefan-Boltzmann.

A modern Nvidia GB200 NVL72 rack dissipates roughly 132 kilowatts. To radiate 100 kilowatts at a radiator surface temperature of 60 degrees Celsius, you need around 160 square meters of radiator. That assumes perfect view to deep space and zero solar absorption. Real designs need two to three times that.

Scale it to a 100 megawatt data center and you are looking at 160,000 square meters of deployable radiator structure. Sixteen hectares of folded engineering that has to unfurl in orbit without binding. The James Webb sunshield was 25 square meters and was the most complex deployment in the history of spaceflight.

You can run the radiator hotter to shrink the area. But the GPU junction temperature caps at roughly 90 degrees Celsius. The physics is the physics.

THE RADIATION PROBLEM

Cosmic rays and solar particles hit silicon. When they do, they cause bit flips, latch-ups, or silent data corruption. Modern GPUs run on 5 nanometer processes. The feature sizes are so small that a single heavy ion can ionize multiple adjacent cells, which defeats simple error correction.

For AI training, the worst flavor is silent data corruption. The system does not crash. The training run produces a model whose weights are quietly wrong. Meta published research in 2022 showing measurable accuracy drift in long training runs from cosmic ray flips at sea level. In low earth orbit the flux is one to two orders of magnitude higher. In the South Atlantic Anomaly it spikes again.

Radiation-hardened silicon exists. The BAE RAD750 powering Mars rovers runs at 200 megahertz and 0.2 watts. You do not train frontier models on that. Modern rad-hard processes lag commercial silicon by ten to fifteen years in performance per watt. You either use commercial Nvidia hardware that gets quietly corrupted in orbit, or you use space-grade silicon that cannot compete on training throughput. Pick one.

THE POWER PROBLEM

Solar in low earth orbit delivers 1361 watts per square meter at the panel face. Modern multi-junction cells convert at roughly 30 percent. For a 100 megawatt data center, accounting for eclipse periods and degradation over a 5 to 10 year service life, you need around 500,000 square meters of solar arrays. Half a square kilometer of deployable structure.

Eclipse storage is the second hit. A typical LEO orbit puts you in Earth’s shadow for 36 minutes out of every 90. To maintain 100 megawatts through eclipse, you need roughly 60 megawatt-hours of battery storage. That is about 350 tons of Tesla Megapack equivalent.

THE INTERCONNECT PROBLEM

This is what actually kills frontier training in space.

Training a modern AI model requires tens of thousands of GPUs synchronizing gradients every few milliseconds. NVLink runs at 900 gigabytes per second per GPU. InfiniBand NDR carries 400 gigabits per second between racks. The training step is synchronous. One slow node stalls the whole batch.

The state of the art for inter-satellite laser links is around 100 to 200 gigabits per second in deployed Starlink V2 systems. Lab demonstrations push toward 1 terabit per second. That is three orders of magnitude below the fabric inside a terrestrial training cluster. Pointing precision, jitter, and atmospheric losses degrade it further.

You cannot do tensor-parallel training across satellites separated by kilometers of vacuum. You cannot do data-parallel training at frontier scale either. The bandwidth simply is not there.

WHAT YOU CAN ACTUALLY DO IN ORBIT

Inference is a different problem. It does not require synchronous coupling across thousands of GPUs. You can pipeline a model across a few satellites, each holding several layers, and serve queries with acceptable latency. Specialized workloads with high power demand and modest compute density work too. Signal processing close to overhead sensors. Sovereign inference for jurisdictions that cannot accept their data crossing certain borders.

That is the actual market. Not training. Not a replacement for Memphis or Iowa. A narrow set of workloads where the value of orbital placement exceeds the order-of-magnitude cost penalty.

WHY THIS MATTERS FOR DEFENSE AND POLICY

The Department of War is buying AI compute in 2026 at a scale the Pentagon has never moved before. Where it places those bets matters. If senior leadership believes orbital data centers are about to give the United States a training advantage in space, the budget and procurement will get pointed at the wrong target.

The honest read is that orbital compute is an inference and sovereignty play. That changes the requirements. You want hardened communications. You want robust pipeline parallelism. You want long-lived satellites with redundant compute payloads. You do not want a million-satellite training megaconstellation. The FCC filing SpaceX submitted in January for a million-satellite orbital AI mesh is a paper exercise that will get right-sized when somebody actually builds a prototype.

The companies positioning themselves for the long orbital play are not trying to put H100s in orbit. They are working on radiation-tolerant inference accelerators, pipeline-parallel architectures, and laser interconnect bandwidth improvements that might, by 2030, make modest in-orbit inference economically defensible for the right customer.

THE BIGGER PRINCIPLE

Every wave of infrastructure hype follows the same pattern. The press release describes the destination. The physics describes the path. The gap between them is where capital gets wasted, retail investors get burned, and policymakers make procurement decisions they regret.

The people who came out ahead on every infrastructure cycle, from railroads to fiber to terrestrial data centers, were not the ones who believed the press release. They were the ones who studied the constraints, understood where the chokepoints actually live, and positioned for the workload that physics permits rather than the one the deck promises.

That is what real AI infrastructure analysis is supposed to do. Strip out the marketing. Do the math. Tell you what is actually shippable.

I break this down daily on my Substack and YouTube. If you want the operational read on AI infrastructure, federal contracting, and the gap between briefed capability and the real thing, follow me everywhere as @vincentactual.

In 1953 he became Chief of Staff of the United States Air Force. In 1957, Chairman of the Joint Chiefs of Staff, the most senior uniformed officer in the entire United States military.

On September 23, 1947, between the combat commands and the Joint Chiefs, Twining was a lieutenant general running the Air Materiel Command at Wright Field in Dayton, Ohio. That day he signed his name to a classified memo and sent it up the chain to the Commanding General of the Army Air Forces in Washington. The subject line read “AMC Opinion Concerning ‘Flying Discs.’”

The operative paragraph:

“It is the opinion that:

a. The phenomenon reported is something real and not visionary or fictitious.

b. There are objects probably approximating the shape of a disc, of such appreciable size as to appear to be as large as man-made aircraft.”

The Twining memo is one of the most cited documents in UAP history. It is almost always treated as a standalone artifact, a single striking quote lifted out of context. It is the opposite of standalone. It is the first move in a documented institutional sequence that ran from September to December 1947, and the famous directive that everyone cites, the one that built private contractors into the foundation of the program, came at the end of that sequence, not the start.

The standard story has the timeline backwards.

My Standing

I spent eight years on active duty in the Navy as an Operations Specialist. Four of those aboard USS Peleliu in TACRON 11 running the Fleet Air Defense Identification Zone in the Pacific. The watch I stood is the same watch the 1947 Air Materiel Command was standing. Identify what’s in the air. Decide friendly, hostile, or unknown. If it’s unknown, escalate. The 1947 packet is what escalation looked like at the strategic level.

What Wright Field Was in 1947

To understand why Twining’s memo carried the weight it did, you have to know what the Air Materiel Command was.

Wright Field was the technical intelligence center of American air power. It was where captured enemy aircraft went to be taken apart. Through 1945 and 1946, under Operation Lusty, the Army Air Forces shipped German jets, rocket engines, radar sets, and airframes to Wright Field for evaluation. The Messerschmitt Me 262. The V-2 components. The Horten flying wings. The German aerospace scientists brought into the country under Project Paperclip were put to work in and around the same facility.

If something flew and the United States wanted to know how, it went to Wright Field. The labs there had spent two solid years reverse-engineering the most advanced aviation technology on earth. That is the institution Twining commanded when he signed the memo. When he wrote that the disc phenomenon was real, he was not a man guessing. He was the man who ran the building where America figured out what was real.

How the Opinion Was Built

Twining’s memo was not a hunch committed to paper. The first paragraph lays out exactly how the opinion was assembled, and the sourcing is the point.

Three streams of data fed it. Interrogation reports furnished by AC/AS-2, the Air Staff intelligence directorate in Washington. Preliminary studies from Twining’s own T-2 intelligence shop. Technical assessment from the Aircraft Laboratory at Engineering Division T-3. The opinion was then finalized in a conference that pulled people from across the Wright Field technical apparatus: the Air Institute of Technology, T-2 Intelligence, the Office of the Chief of Engineering Division, and the Aircraft, Power Plant, and Propeller Laboratories.

That list is every technical exploitation directorate the Army Air Forces had at Wright Field. The same labs that dismantled the Me 262. The same engineers who studied the V-2. Twining was telling Washington that the entire apparatus had looked at the data and reached a consensus. This was not one officer’s read passed up the chain. It was the institution’s finding.

What the Objects Were Doing

Here is the part of the memo that almost never gets quoted, and it is the most concrete evidence in the entire 1947 packet. After stating that the phenomenon was real, Twining’s memo describes what the objects actually did, drawn from the interrogation reports.

Extreme rates of climb. Maneuverability, particularly in roll. Action that must be considered evasive when the objects were sighted or contacted by friendly aircraft and radar. No trail, except in a few instances when the object was apparently operating under high performance. Circular or elliptical in shape, flat on the bottom, domed on top. Formation flights of three to nine objects. Normally no sound, except three instances of a substantial rumbling roar. Level flight speeds normally estimated above 300 knots.

That is not a description of weather balloons or misidentified aircraft. It is a target track. Evasive action when contacted by radar is a behavior, not an optical illusion. Formation flights of three to nine objects is a count, made by trained observers. The man who ran the technical intelligence center of the United States Air Force read those reports, convened his laboratories, and signed his name to the conclusion that the objects were real.

Twining then did the engineering math. Could American industry build something like this? His answer was yes, in principle, with extensive detailed development, capable of roughly 7,000 miles range at subsonic speed. But any such program would be expensive, slow, and would have to run independently of existing projects. And then the line that should stop anyone reading the memo:

“Due consideration must be given the following: (1) The possibility that these objects are of domestic origin, the product of some high security project.”

A lieutenant general writing to the Commanding General of the Army Air Forces, on classified stationery, raising the possibility that the objects were an American black program he had not been read into. He closes the memo by requesting “a complete interchange of data.” Twining is asking Washington to set up a coordinated investigation. He does not yet have the authority to stand one up himself.

That authority would come three months later, in the Craigie directive. The directive was the answer to this memo.

One Day Later: The Loedding Drawing

The day after Twining signed his opinion, a different office at Wright Field sent a technical drawing up the same chain.

The drawing was titled “Loedding Flying Disc,” designated LD-2. Alfred Loedding was a civilian aeronautical engineer in T-2 at Wright Field, a man with a documented professional interest in low-aspect-ratio and disc-form aircraft. He would go on to become one of the central technical figures in Project Sign. The drawing was being routed up because Major General George McDonald, the Director of Intelligence at AC/AS-2, had asked for it by name.

The cover memo carried one specific instruction: “Because of patent rights involved, it is requested that a record be kept of all persons reviewing this drawing by witnessing with signatures and dates directly on the drawing.”

Patent rights. On a classified drawing of a flying disc. Tracked by signature.

The detail matters because of what it tells you about the office. The same T-2 shop that had just contributed to the opinion that the unexplained disc phenomenon was real was, the next day, routing a classified American disc design with its own intellectual property concerns. The investigation of disc-form objects and the development of disc-form aircraft were not in separate buildings or separate programs. They were the same people, the same week.

The same memo also carried a foreign intelligence note. A U.S. Military Attaché report out of Moscow, dated 9 June 1947, claimed the Soviets were building 1,600 aircraft based directly or indirectly on the German Horten VIII flying-wing design, a 131-foot-wingspan machine, jet-propelled in the Russian version, intended for bomber squadrons. Whether or not that number was accurate, it tells you what Wright Field was worried about. Disc and wing forms, at scale, in Soviet hands. The unexplained phenomenon was being evaluated in that exact threat context.

October: The Phenomenon Gets a File Number

By October 30, 1947, the topic had a permanent home in the filing system. Air Force Headquarters opened index entry 0009, subject “Flying Saucer - Phenomena,” filed under decimal 350.09. The summary line read “Intelligence Requirements on Flying Saucer Type Aircraft.”

The Army decimal filing system was not arbitrary. The 350 series was Intelligence. It held foreign capability assessments, order of battle reports, attaché traffic. Assigning the phenomenon a 350.09 number with a standing “intelligence requirements” heading meant the Air Staff had reclassified it from a news event into a recurring intelligence target, the same bureaucratic category as tracking a foreign weapons program.

Bureaucracies do not open permanent file numbers for things they expect to go away. Ninety days after the summer 1947 sightings, the phenomenon had a file number and a requirements heading. It was now part of the permanent intelligence workload.

December 19: McCoy Sends the Package Up

By the third week of December, enough material had accumulated at Wright Field to justify a formal transmittal to the top. On December 19, 1947, Colonel H.M. McCoy, Twining’s Chief of Intelligence at the Air Materiel Command, sent a packet of reports directly to the Chief of Staff of the United States Air Force.

McCoy’s cover note is one paragraph, and every clause in it is doing work:

“Continued and recent reports from qualified observers concerning this phenomenon still makes this matter one of concern to Headquarters, Air Materiel Command. Intelligence Department of this Command is continuing the collection and analysis of all available reports.”

Continued and recent reports. Not a closed case. Qualified observers. Not cranks. Continuing the collection and analysis. An active, ongoing intelligence operation, described in the present tense, by the colonel running it, addressed to the most senior officer in the Air Force. The packet underneath that cover note held the Twining memo, the Loedding material, the file index, AMC’s monitoring of newspaper coverage, and more. Three months of work, bundled and sent to the top.

December 22: The Air Force Says “Film Defects”

Three days after McCoy’s package reached Washington, a separate piece of correspondence went the other direction, back down the chain to California.

The subject was a set of photographs. In November 1946, Mrs. Mary L. Herren had taken pictures near Jefferson, Oregon, that contained marks she thought might be flying discs. The photographs worked their way to the Fourth Air Force intelligence office at Hamilton Field, California, where the intelligence officer, Donald L. Springer, took them seriously enough to forward them to Washington with a note describing Mrs. Herren as “a very reliable source of information.”

This is the reply that came back, signed by a lieutenant colonel for the Chief of Staff:

“The marks appearing on the negatives enclosed in basic letter are believed to be defects in the film, paper, or camera and not pictures of ‘flying discs.’

It is requested that no further investigation be made of this incident.”

No analysis was attached. No lab at Wright Field is named as having examined the negatives. The same technical apparatus that three months earlier had convened its laboratories to produce the Twining opinion was not applied here at all. The marks were declared film defects, and the order was to stop.

Hold the two documents side by side. December 19, the institution’s internal record: an active intelligence program, qualified observers, real phenomenon, sent to the Chief of Staff. December 22, the institution’s external posture: film defects, no further investigation, signed for the Chief of Staff. Same office. Same week. Two completely different documents, and the public would only ever have been positioned to see the second kind.

December 30: Craigie Makes It Permanent

Eight days after the Hamilton Field brush-off, Major General L.C. Craigie, the Director of Research and Development, acting by command of the Chief of Staff of the now-independent United States Air Force, issued the directive that ordered the Air Materiel Command to stand up the project that became Project Sign.

The operative paragraph instructed the new project to “collect, collate, evaluate and distribute to interested government agencies and Contractors all information concerning sightings and phenomena in the atmosphere.” The capitalization of “Contractors” is in the original government document. It is not a transcription choice.

That clause is the structural seed of every UAP classification fight that has happened since. It put private contractors inside the distribution loop of the program on day one, by directive, as standard procedure. Seventy-eight years later, in 2023, the Schumer-Rounds amendment tried to write Congress a tool to reach into exactly that contractor space. The amendment died in conference committee with no recorded opposition. The 1947 directive is still operative.

But notice where the directive sits. It is dated December 30. It is the seventh document in a sequence that opened on September 23 with Twining writing “real and not visionary or fictitious.” Craigie did not start the program. He formalized one that had already been running for three months.

The Sequence, Start to Finish

Lay the seven documents out by date and the shape is unmistakable.

September 23: Twining’s laboratories deliver a formal opinion that the phenomenon is real, with a physical description and an engineering assessment. September 24: the same intelligence shop routes a classified American disc design with a foreign threat note attached. October 30: the Air Staff opens a permanent intelligence file number. Through the fall: the Air Materiel Command monitors press coverage and foreign claims. December 19: the Chief of Intelligence at AMC tells the Chief of Staff in writing that the matter remains active and under analysis. December 22: Air Force Headquarters disposes of a civilian photographic case as film defects with no analysis. December 30: the contractor-distribution directive is signed.

That is not a series of reactive decisions made under pressure. It is a program being assembled, step by step, in under four months. By the time Craigie signed the famous directive, the institution had already decided the phenomenon was real, given it a file number, set up its monitoring, and worked out its public posture.

The standard disclosure narrative assumes the government has spent nearly eight decades trying to figure out what to do about UAP reports. The 1947 packet shows the handling questions were settled in the fall of 1947. What has happened since is not deliberation. It is continuity.

Why This Matters Now

The constitutional question is not about flying discs. It is about whether the architecture the Air Force stood up in 1947, later formalized into permanent operation under the Atomic Energy Act framework and extended through the contractor special access program system, has produced an oversight gap that exceeds what Article I of the Constitution contemplates.

That question is legislatively actionable. It does not require believing anything in particular about UAP. It requires reading the documented chain of institutional decisions from September to December 1947 and asking whether Congress today has the legal reach to compel disclosure from the structure that chain built.

The man who signed the founding opinion went on to chair the Joint Chiefs of Staff. The directive that put contractors in the loop has never been rescinded. The file number is older than almost everyone reading this. Seventy-eight years is enough time to evaluate the result.

Stay sharp.

That sentence appears in paragraph two of a SHAEF Air Staff memo dated 11 February 1945. The classification is Top Secret. The reference number is SHAEF AIR/TS.37153/A-2. The signer is Air Commodore C.M. Grierson, Assistant Chief of Staff for Intelligence at Supreme Headquarters Allied Expeditionary Force. The subject line reads “Night Phenomena.”

Mr. Robertson is Howard Percy Robertson, the Princeton physicist serving as head of the Scientific Investigation Division at SHAEF. He is the same Howard Robertson who, eight years and three months later, would chair the CIA-convened panel of scientists that recommended the United States government adopt a policy of public debunking on the subject of unidentified flying objects. The Robertson Panel of January 1953 is the foundational document of the modern disclosure-suppression regime. The Grierson memo of February 1945 is the moment Robertson was first read into the problem.

Yesterday’s piece established the constitutional architecture that the Atomic Energy Act of 1946 made permanent. Today’s piece shows that the architecture was already operational eighteen months before the AEA was signed, and that the man who would design its public-facing debunking strategy was personally inside the program in 1945.

Standing

Eight years on active duty in the Navy as an Operations Specialist. Four of those years aboard the USS Peleliu running the Fleet Air Defense Identification Zone in the Pacific. Tracking unknowns over a strike group is a watch you stand. You decide friendly, hostile, or unknown. If it’s unknown, you escalate.

The men flying with the 415th Night Fighter Squadron in December 1944 stood the same watch in a different theater. P-61 Black Widow crews, operating out of bases in Alsace, intercepting Luftwaffe night raids and German V-weapon launches. The job was identifying every contact in the air around them and deciding what it was. When they encountered something they could not identify, they wrote it up.

What they wrote up in December 1944 became the file we have today.

The File

The file index is SHAEF AIR/TS.37153/A-2. “TS” is Top Secret. “37153” is the subject file number for the matter at hand. “A-2” is the Air Section of the Assistant Chief of Staff for Intelligence at Supreme Headquarters Allied Expeditionary Force.

A-2 is the same intelligence shop that handled order of battle, photographic reconnaissance assessments, Luftwaffe technical intelligence, and German weapons system recovery operations. The Night Phenomena file sat inside the same Top Secret registry as the V-2 rocket intelligence and the jet propulsion technical exploitation files. Same office, same drawer, same classification.

That detail matters. The phenomenon was not being filed in some marginal category for crank reports. It was being filed in the highest-classification intelligence registry of the Allied command, in the same registry that held the most sensitive German weapons technology files.

What the 415th Reported

The original handoff from XII Tactical Air Command into the SHAEF chain quotes the 415th’s training and tactical information for December 1944 verbatim. The squadron’s intelligence officer reproduced the language that would define everything that came after.

The squadron’s intelligence officer, Captain B. Ringwald, transmitted the sortie report extracts to higher headquarters on 30 January 1945. Specific dates, specific altitudes, specific behaviors.

Night of 14-15 December 1944: “In vicinity of Erstein flying at 1000 ft. observed large red light at 2000 ft. going East at 18:40 hrs. Travelling at approximately 200 MPH.”

Night of 16-17 December 1944: “20 miles North of Breisach at 800 ft. observed 5 or 6 flashing red and green lights in ‘T’ shape. Thought they were flak. About 10 minutes later saw the same lights much closer and behind me. We turned port and starboard and the lights followed.”

Night of 22-23 December 1944: “Two lights coming towards A/C from the ground. Upon reaching altitude of plane, they leveled off and stayed on my tail for approximately 2 minutes. Lights appeared to be a large orange glow. After staying with A/C for approximately 2 minutes, they would peel off and turn away, fly along level for a few minutes and then go out. They appeared to be under perfect control at all times.”

Night of 29-30 January 1945: “Sighted a Foofighter about half way between Weissembourg and Landau. Foofighter was off to the starboard and rear at Angels 2. Lights were amber and one was 20-50 ft. above the other and of about 30 seconds duration. Foofighter was about 1000 ft. away and following.”

Captain Ringwald included a footnote at the end of the report. “Foofighters is the name given these phenomenon by combat crews of this Squadron.”

The pilots gave the things a name. The name made it into the official record. The lights followed aircraft. They flew formation. They matched maneuvers. They appeared, in the words of the sortie report, “under perfect control at all times.”

In every case, the pilot called GCI Control and asked if there was a friendly or hostile bogey in the area. In every case, GCI Control returned a negative answer.

The Chain of Distribution

The 415th’s report went to XII Tactical Air Command. From there to First Tactical Air Force Provisional. From there to Air Staff SHAEF. From SHAEF to the Air Ministry in Whitehall. From the Air Ministry to A.D.I.(K), the Air Intelligence directorate that handled prisoner-of-war interrogation. From A.D.I.(K) back to SHAEF.

Reverse copies went to United States Strategic Air Forces in Europe, Ninth Air Force, Bomber Command, and Second Tactical Air Force. Every operational air command in the European Theater had a copy of the Night Phenomena file by March 1945.

The Air Ministry’s reply, dated 13 March 1945 and signed by Group Captain Hopkins, conceded that “Bomber Command crews have for some time been reporting similar phenomena” and stated that “the whole affair is still something of a mystery and the evidence is very sketchy and varied so that no definite and satisfactory explanation can yet be given.”

Group Captain Hopkins offered two candidate explanations. The first was that some of the phenomena might be Messerschmitt Me 262 jets. The second was that they might be German flak rockets. Neither explanation accounted for the controlled-flight characteristics in the 415th’s reports.

The Grierson Memo

The single most important document in the file is the one Grierson signed on 11 February 1945. It is reproduced in full below.

The operative passage is paragraph one. Read it carefully.

“From the number of reports quoted in the 2nd W/Ind. from the 415th Night Fighter Squadron, it would seem that there must be something more than mere imagination behind the matter, and in view of the fact that pilots and crews are becoming slightly worried by them, it is considered that everything possible should be done to get to the root of the matter.”

A flag officer in the Allied air command, writing on Top Secret stationery in February 1945, states on the record that the phenomenon is real, the aircrews are taking it seriously, and the institution intends to “get to the root of the matter.”

Paragraph two routes the problem to two destinations: the Air Ministry, and “the Scientific Investigation Division of this Headquarters (Mr. Robertson).”

Howard Percy Robertson, Princeton physicist, then serving as Eisenhower’s scientific advisor at SHAEF. The man who would, eight years later, sit at the head of the CIA-organized Robertson Panel and recommend that the US government adopt a policy of public debunking, news media engagement, and educational counter-programming to suppress public interest in unidentified flying objects.

The Robertson Connection

The 1953 Robertson Panel produced a report whose key recommendation reads as follows. The panel “concluded that the continued emphasis on the reporting of these phenomena does, in these perilous times, result in a threat to the orderly functioning of the protective organs of the body politic.” It recommended that the National Security Council take steps “to strip the Unidentified Flying Objects of the special status they have been given and the aura of mystery they have unfortunately acquired.”

The man who chaired that panel had been working the same problem on Eisenhower’s staff in 1945. Howard Robertson did not come to UFO debunking fresh in January 1953. He had eight years of prior exposure to the data, starting with the Grierson memo. By the time he was asked to convene the CIA panel, he had personally seen what was in the SHAEF Top Secret registry on Night Phenomena.

This matters for the constitutional argument. The Robertson Panel is often cited as the moment the United States government formalized its public-facing posture on UAP. That posture has been broadly consistent for seventy-three years. What the Grierson memo shows is that the man who designed the public-facing posture had been read into the underlying classified investigation for eight years before he convened the panel.

The 1953 recommendation was not the conclusion of a scientific inquiry. It was the public-facing handling decision made by an insider.

Continuity Through the AEA

The Atomic Energy Act of 1946 created a permanent classification regime that operates outside Congressional oversight and human classification authority. Yesterday’s piece walked through what the AEA enables structurally.

The SHAEF Air/TS.37153/A-2 file shows that the architecture the AEA later codified was already functionally operational in early 1945. The chain of custody for unexplained-aerial-phenomena reports was already running. The classification posture was already Top Secret from inception. The scientific advisory channel was already routing through one man.

The AEA didn’t invent the architecture. The AEA gave the architecture statutory permanence after the war. The personnel who had been running the wartime version transitioned into the postwar version with minimal break.

Howard Robertson stayed on. He moved from SHAEF Scientific Investigation Division to senior advisory roles at the Department of Defense and the CIA. The intelligence personnel who handled the 1945 file moved into the Air Materiel Command intelligence apparatus at Wright Field that would, in September 1947, issue the Twining memo confirming that “the phenomenon reported is something real and not visionary or fictitious.”

Twining was working the same problem the SHAEF A-2 office had been working in 1945. Same problem, same continuity of personnel, two years apart, with the Atomic Energy Act passed in between to make the handling permanent.

What This Changes

The standard disclosure narrative starts in 1947 with Roswell or the Arnold sighting. The Grierson memo of February 1945 establishes a different starting point. The Allied air command was already running a classified, structured, multi-echelon investigation of unexplained aerial phenomena two and a half years before Roswell. The same scientist who would later design the public-facing debunking strategy was personally inside the wartime investigation.

When the AEA passed in 1946, it did not start the program. It made an existing program permanent.

When the Robertson Panel recommended public debunking in 1953, it was not the conclusion of an external investigation. It was the public-facing handling decision made by the same individual who had been inside the classified investigation since 1945.

The disclosure debate proceeds as if the institutional posture is the result of seventy years of careful inquiry that has produced no findings. The Grierson memo is one piece of evidence that the institutional posture is older than that, was decided early, and has been carried forward by continuity of personnel from inside the original wartime program.

Tomorrow

Tomorrow’s piece walks through the full twenty-eight page packet from Air Materiel Command in 1946 and 1947. The signed Twining memo confirming the phenomenon was “real and not visionary or fictitious.” The administrative routing slips showing how the postwar program was wired into the same intelligence apparatus that had handled the 1945 file. The pieces are continuous.

The architecture didn’t start in 1947. The architecture didn’t start in 1946. The architecture was already running in 1945, on Top Secret SHAEF letterhead, with the man who would later run the public-facing debunking strategy already inside.

Stay sharp.

In 2023, the Senate Majority Leader introduced an amendment to the National Defense Authorization Act that would have done two specific things. It would have invoked federal eminent domain to compel the return of any non-human technology held by private contractors. And it would have created a review board with subpoena power capable of piercing contractor-held special access programs. The amendment had bipartisan sponsorship from Senator Mike Rounds. It cleared the Senate. Then it died in conference committee.

Think about what that means. The Senate Majority Leader, working with bipartisan support, in his own chamber, on his own amendment, in the bill that funds the Department of Defense, could not keep the language alive long enough to reach the President’s desk. Something stronger than partisan politics killed it, because partisan politics doesn’t usually defeat the Majority Leader on his home turf. The question this piece starts with is what does.

The answer involves a 1946 law that most Americans haven’t thought about since high school civics, if they ever did. The Atomic Energy Act of 1946 built a classification architecture that operates outside the framework Congress thinks it’s exercising oversight over. The structure has been inherited, extended, and shielded by every administration since Truman, and it is the reason disclosure stalls regardless of which party holds power, which president signs which executive order, or which whistleblower goes on which podcast. The disclosure debate as currently framed is mostly the wrong debate. The right one is constitutional.

Standing

I spent eight years on active duty in the Navy as an Operations Specialist, with four of those years aboard the USS Peleliu in TACRON 11 running the Fleet Air Defense Identification Zone. My job was tracking what was in the air around the strike group and deciding whether it was friendly, hostile, or unknown. Off the West Coast in the Pacific, my watch team logged contacts that didn’t behave like anything in any manual we’d been issued.

This article is not about that. It is not about whether the official explanation of what we saw is correct. I’m using that background to earn a few paragraphs of your attention, and I’m spending that attention on a question that does not require you to believe any specific claim about what UAP are. The question is whether the United States has built a classification regime that exceeds what Article I of the Constitution contemplates. That question is actionable through legislation. The metaphysical question is not.

The 1946 Problem

Most discussions of government secrecy in the United States focus on the Department of Defense Special Access Program framework or the Director of National Intelligence’s compartmented control system. Both of those are real, both are powerful, and both have been written about extensively. Neither of them is the most secret classification system in the federal government.

The Atomic Energy Act of 1946 created a parallel classification regime that sits outside the DoD SAP world and outside the IC compartmentation world. Under the AEA, certain categories of information are born classified by operation of law. The information doesn’t have to be reviewed and stamped by a classification authority. It is classified the moment it exists. That is a unique authority. Every other classification regime in the federal government requires a human being to make a classification decision. The AEA does not.

The Atomic Energy Commission, the original AEA implementing body, also created its own clearance system separate from the DoD’s. Q clearance is administered today by the Department of Energy, not by the DoD or the IC, and the standards for Q access are stricter than Top Secret. Q clearance covers Restricted Data and Formerly Restricted Data, categories that exist nowhere else in federal classification practice. Q-cleared programs can be compartmented in ways that don’t appear in DoD SAP registries, because they aren’t DoD SAPs. They are DOE programs, and DOE answers to a different oversight pipeline than the Pentagon does.

When the AEC was abolished in 1974, its authority transferred to the Energy Research and Development Administration. When ERDA was abolished in 1977, its authority transferred to the Department of Energy. None of the underlying classification authority was diminished in either transition. The AEA classification regime is intact and operating today, eighty years after it was written, and it covers everything the DOE national laboratory system does.

The point is worth restating before going further. There is a classification system inside the United States government that operates under stricter rules than the SAP framework, doesn’t require human review to classify information, and is administered by a department that most Americans don’t associate with national security. It was created in 1946 to protect the secret of the atomic bomb, and it has been continuously extended ever since.

The Cover Structure

If you wanted to hide a long-running classified research program in the most legally defensible way available to the executive branch, you wouldn’t put it in a DoD SAP. DoD SAPs have congressional reporting requirements, even if those requirements are weaker than the public thinks. The Gang of Eight has at least nominal access. Inspector General offices have at least nominal jurisdiction. Whistleblower statutes apply with at least nominal force.

You would put it in DOE. Specifically, you would put it in the national laboratory system.

The national labs are formally civilian research institutions, operated by federal contractors under contract to DOE. Sandia National Laboratories is operated by National Technology and Engineering Solutions of Sandia, a Honeywell subsidiary. Los Alamos National Laboratory is operated by Triad National Security, a consortium of the University of California, Battelle, and Texas A&M. Lawrence Livermore National Laboratory is operated by Lawrence Livermore National Security, a consortium of the University of California and Bechtel. These are not government employees. They are contractor employees holding Q clearances and working on contract-classified research at facilities the federal government owns but does not directly staff.

That structure accomplishes several things at once. It puts academic and civilian cover over the underlying work, which makes the research harder for outsiders to identify as classified. It gives the program federal contractor protections against the Freedom of Information Act, since contractor-held documents fall outside FOIA’s reach in ways government-held documents do not. It runs the personnel through a clearance pipeline that DoD and the IC do not control. And it allows funding to flow through DOE appropriations lines that look on paper like university research grants and infrastructure maintenance, not classified weapons development.

This is the architecture. Not a bunker. Not a black site. A statutory framework, a contractor structure, and a budget pipeline, all operating together inside the lawful federal research apparatus.

The 1947 Receipt

The contractor-cover architecture isn’t a theoretical inference from how the AEA was written. There is a signed military order establishing it.

On December 30, 1947, Major General L. C. Craigie, Director of Research and Development, acting by command of the Chief of Staff of the newly-independent United States Air Force, directed the Commanding General of Air Materiel Command to set up the project that became Project Sign, the Air Force’s first formal UAP investigation. The operative paragraph is one sentence long, and it is worth reading carefully:

“In implementing this policy, it is desired that the Air Materiel Command set up a project whose purpose is to collect, collate, evaluate and distribute to interested government agencies and Contractors all information concerning sightings and phenomena in the atmosphere which can be construed to be of concern to the national security.”

December 30, 1947. Seventeen months after the Atomic Energy Act was signed. The founding charter of the Air Force’s first UAP investigation authorizes distribution to private contractors as a core operating principle, not as an exception. The capitalization of “Contractors” in the original is the government’s, not mine.

The directive doesn’t define which contractors. It doesn’t require congressional notification of which contractors are included. It doesn’t impose audit requirements on what contractors do with the data once they receive it. It just builds the pipe. And the pipe has been there ever since.

When you read the Schumer-Rounds amendment of 2023 against this 1947 document, the structural continuity is obvious. The amendment was trying to pierce a contractor architecture whose authority to exist was written into Air Force policy seventy-six years earlier and has never been rescinded.

The Manhattan Comparison

The Manhattan Project is the obvious historical reference point and it is worth understanding why the comparison is more structural than people assume.

Manhattan was the largest classified program in American history at the time. It employed roughly 130,000 people across multiple sites. It cost approximately thirty billion dollars in today’s money. And a meaningful percentage of the people working on it didn’t know what they were building. Most participants did component work without seeing the whole. That compartmentation model is exactly what you would need to run a long-duration classified research program at scale, and it is the model the AEC inherited from Manhattan and that DOE inherited from the AEC.

Compartmentation at the Manhattan scale wasn’t invented for Manhattan. The intellectual tools came from older intelligence and counterintelligence practice. What Manhattan did was industrialize it. It demonstrated that you could employ a hundred thousand people on a single classified objective and prevent leaks across the workforce indefinitely, as long as the legal authority for the compartmentation held. The AEA gave that authority statutory permanence. DOE has been running the model continuously for eighty years.

There is one critical difference between Manhattan and what the AEA-DOE structure enables today. Manhattan was a wartime program with explicit congressional appropriation and a defined end state. The bomb was finished, the war ended, and the program was wound down into a peacetime regulatory body. The current architecture has no end state. It has no public objective. It has no defined sunset. The classification authority that sustains it is permanent unless Congress repeals or amends the AEA, and Congress has shown no appetite for either move in eighty years.

The Personnel Pattern

If a long-running compartmented program exists inside this architecture, the people who would have run it would have specific career signatures. They would be senior Air Force Research Laboratory or DOE-adjacent technical leaders. They would have worked the Sandia-Kirtland-Los Alamos triangle in New Mexico, where the national lab system, the Air Force Materiel Command intelligence apparatus, and the contractor ecosystem overlap most densely. They would have retired into private aerospace and applied technology firms that subcontract back into that same ecosystem.

That pattern is worth tracking. James McCasland, who commanded the Air Force Research Laboratory at Wright-Patterson, retired to New Mexico and joined Applied Technology Associates, a firm with deep Sandia-Kirtland-Los Alamos subcontracting exposure. He is the highest-profile name in this pattern but not the only one. Eskridge in Huntsville sits in the rocket and missile defense ecosystem that interfaces with DOE through the National Nuclear Security Administration. Names like Chavez and Casillas appear at Los Alamos in roles that touch the same program areas. None of this proves anything on its own. What it shows is that the personnel are clustering exactly where the AEA-DOE architecture is densest, and that pattern is not random.

I am not naming these people to accuse them of anything. I am noting that if you wanted to find the workforce of a contractor-held long-running classified program operating under DOE cover, this is precisely the personnel pattern you would expect to see, and this is precisely the personnel pattern that shows up in the public record.

The Wilson-Davis Memo

There is a specific document worth knowing about if you don’t already. The Wilson-Davis memo is allegedly a set of notes from a 2002 meeting between Admiral Thomas Wilson, then Deputy Director of the Defense Intelligence Agency, and astrophysicist Eric Davis. The memo describes Wilson attempting to get read into a Special Access Program related to crash retrieval material and being denied access by a private aerospace contractor’s security officer, who told him the program had been moved out of normal DoD oversight.

The authenticity of the memo is contested. Wilson has neither confirmed nor forcefully denied it. Davis has confirmed the meeting occurred. Without independent verification of the memo’s contents, it cannot be treated as proven evidence. It is, however, consistent with the structural argument this article is making. The Wilson-Davis memo, if accurate, describes a program that has migrated from DoD-held to contractor-held status, with funding flowing through channels that aren’t subject to standard congressional oversight. That is the AEA-DOE-contractor structure described above, and that is what the Schumer amendment in 2023 was specifically trying to pierce.

I’ll cover the Wilson-Davis memo and its authenticity questions at greater length in a separate piece. For purposes of this article, the relevant point is that the structure the memo describes is the structure the AEA enables, regardless of whether the specific memo is real.

What the Schumer Amendment Would Have Done

Return to the Schumer-Rounds amendment. The specific provisions of the amendment matter, because they tell you what the drafters thought was wrong with the current structure.

The amendment would have invoked federal eminent domain against private contractors holding non-human technology. That provision only makes sense if you believe the technology, if it exists, is held in private contractor custody under arrangements that block normal government recovery. That is the AEA-DOE-contractor structure.

The amendment would have created a review board with subpoena power capable of piercing contractor-held SAPs. That provision only makes sense if you believe existing oversight bodies, including the Gang of Eight and standing inspectors general, lack the legal reach to compel disclosure from contractor-held programs. That is the AEA-DOE-contractor structure.

The amendment was gutted in conference committee. Conference committee is a closed-door process. The conferees who removed the language did not have to publicly defend their position. The vote that ratified the gutted final text was a vote on the entire NDAA, which is a bill no senator wants to be seen voting against. The political cost of killing the amendment was therefore essentially zero, and the political cost of letting it pass would have been substantial for whoever the program structure protects.

The fact that the amendment was killed quietly and without recorded opposition tells you the resistance is institutional, not partisan. The resistance came from somewhere. The somewhere is consistent with what an actively operating AEA-DOE-contractor program would defend.

The Constitutional Stakes

Pull the threads together.

The Atomic Energy Act of 1946 created a classification regime that operates without human review, under stricter rules than DoD SAPs, administered by a department most Americans don’t associate with national security. That regime survives every administration. It is operated through a contractor structure that puts academic and civilian cover over the underlying work, blocks FOIA, runs personnel through a separate clearance pipeline, and routes funding through DOE appropriations that look like research grants. The personnel pattern of retired senior leaders concentrates in the geographic and contractual zones where this architecture is densest. The one congressional attempt in eighty years to pierce contractor-held SAPs under this structure was killed in conference committee with no public opposition. No public unveiling is planned. Compartmentation is strict enough that even cleared personnel at the SAPCO level can be denied access. The program structure, if a program is in fact operating inside it, is designed to outlast presidential administrations indefinitely.

That is not a bigger Manhattan Project. It is a parallel branch of the executive operating on its own legal authority and accountable to no one outside its own chain. Whether or not anything non-human is being studied inside it is the wrong question. The framework that would be required to study it has constitutional implications on its own terms.

What to Take From This

Three things.

First, the disclosure debate as it is currently framed is mostly the wrong debate. Whether the bodies are real and whether the craft are alien are interesting questions, but they are not the questions Congress can act on. The constitutional question is whether the AEA classification regime and the contractor-held SAP framework have produced an oversight gap that exceeds what Article I contemplates. That question is actionable. The metaphysical question is not.

Second, the cases that keep producing strange institutional behavior cluster in the parts of the federal research apparatus that operate under AEA authorities and contractor cover. That is a pattern worth tracking on its own merits, regardless of what you believe about UAP.

Third, the structural reform that would matter is amending the Atomic Energy Act and creating a contractor SAP review authority with real subpoena power. The Schumer amendment failure is a data point on how hard that fight is. It is not a reason to stop fighting it.

You don’t need aliens for any of this to matter. The 1946 problem is a problem on its own terms, and it is one that congressional staff and national security reform advocates should be talking about regardless of how the disclosure question resolves.

I’ll keep covering this. Coming pieces will look at the Wilson-Davis memo and what it does and doesn’t tell us about the contractor structure, what the Schumer amendment would have actually changed if it had survived conference, and what the McCasland case looks like when you read it through the AEA framework instead of the disclosure framework. The piece after that will work through a 28-page packet of declassified Air Materiel Command correspondence from 1946 and 1947 that shows the operational receipt of this architecture being built in real time.

Stay sharp.

Empty Waymos sitting on residential side streets. Idle vehicles cycling the same blocks waiting for a dispatch. A stopped robotaxi froze traffic on the Venetian Causeway and went viral before the public launch even happened.

This is not a glitch. It is the economic model.

THE DEADHEAD PROBLEM

When a Waymo finishes a paid ride and does not have a next pickup queued, it has to go somewhere. It cannot sit in an active travel lane. Parking garages cost money. So the fleet algorithm sends it to cruise residential streets or stage in low-enforcement zones until the next dispatch lands.

Those are deadhead miles. Empty trips. No passenger, no revenue, just a vehicle burning electricity and adding to traffic so the company does not lose its position in the dispatch queue.

Every minute a Waymo sits in a paid garage costs Alphabet money. Every minute it cruises a residential block, the externality lands on you.

THE HURRICANE SCENARIO IS NOT HYPOTHETICAL

San Francisco already showed what happens when the grid drops. In December 2025, a mass power outage took out traffic lights across the city. Waymos stopped in the middle of intersections. The fleet algorithm could not parse a four-way stop without functioning signals, so the cars froze where they were. San Francisco’s mayor reported stalled Waymos blocking emergency vehicles. The company suspended service citywide. A city supervisor called Waymo executives to a hearing.

That was a clear-sky outage. No storm. No surge. No evacuation order. Just traffic lights going dark for a few hours.

Now run the Miami Beach version. A Category 3 hurricane is forty-eight hours out. Mandatory evacuation is in effect for the barrier islands. FPL has already cut power to flood-prone feeders as a precaution. Lincoln Road and Collins Avenue signals are dark. The MacArthur, Venetian, and Julia Tuttle are running outbound at capacity with thousands of residents trying to clear the island before the bridges close.

A Waymo finishing its last paid ride hits a dark intersection. The fleet algorithm pauses. The car stops in the middle of the lane.

Multiply that by the fleet size. Every Waymo that was on the island when the grid went down is now sitting in an intersection or a causeway approach lane, because the failure mode is not crash. It is freeze. There is no driver to override. No remote operator with the latency to clear dozens of vehicles fast enough. No tow capacity prestaged to physically remove a 4,800 pound stalled vehicle in time for the evacuation window to hold.

San Francisco got a hearing. Miami Beach gets bodies in cars on a causeway in a Cat 3.

The fleet is operating on the assumption that the grid is up and the cell network is reachable. Both assumptions break first in a hurricane. The vehicles are designed to fail safe, which on this island means fail on the only routes off it.

The commission should be demanding the exact failure-mode protocol before the fleet doubles. What happens when traffic signals go dark across a neighborhood. What happens when the cell network degrades. What is the maximum number of vehicles Waymo can remotely clear per hour during a regional emergency. Where is Moove staging tow capacity for stalled fleet vehicles. Who pays when a frozen robotaxi delays a fire truck or an ambulance. None of this is in the current permit conditions, because the conditions were written for a normal-weather operating model.

Atlantic hurricane season opens June 1.

SAN FRANCISCO ALREADY RAN THIS PLAY

Waymo collected 589 parking tickets in San Francisco in 2024. The company paid every one of them as a line item.

That number sounds bad until you do the math. A San Francisco parking ticket runs roughly $80 to $120. Call it $100,000 a year in tickets. Waymo was clearing 250,000 paid rides per week in San Francisco at peak. The ticket cost was less than a rounding error on revenue.

Residents in the Sunset and Richmond started documenting empty Waymos staging on their blocks at 3am. Sleep disruption from headlights and idle hum. Parking spots taken by vehicles with no human inside. The SFMTA had to write tickets aggressively just to push the fleet out of residential zones, and even then Waymo only built dedicated staging lots after the city made it expensive enough.

The lesson is simple. The ticket is a feature, not a bug. The company absorbs it. The residents lose the parking space anyway.

THE EXTERNALITIES THE TICKETS DON’T PRICE

The 589 parking tickets are the visible cost. They are also the smallest line item in what the fleet actually extracts from the city.

Curb space is a public good. Every block of residential parking was built and maintained on the property tax base. When a Waymo stages on that curb overnight, the city is renting capacity to Alphabet at zero. The resident who pays the property tax loses the spot. The company that absorbs the ticket captures the value. That is a transfer of public infrastructure to a private balance sheet, and it is happening at scale.

Sleep is the next one. A Jaguar I-Pace running idle systems is not silent. Cooling fans cycle. The LiDAR units, the spinning laser scanners on the roof that map the surroundings, rotate constantly. Headlights wash the front of the house when the fleet algorithm pings the vehicle awake at 3am to relocate. None of that shows up on a parking ticket. The cost lands on whoever sleeps in the front bedroom and never sees a dollar of compensation.

Empty miles eat road capacity, signal cycles, and causeway throughput that paying drivers compete for the rest of the day. Waymo does not reimburse the city for any of it. Not for asphalt wear, not for traffic signal maintenance, not for the time other drivers lose stuck behind a cruising empty car. Transportation planners measure this drag under vehicle miles traveled, the metric that drives everything from road budgets to air quality models. Studies of ride-hail fleets before robotaxis arrived already pegged deadheading at 20 to 40 percent of total miles on the network. Autonomous fleets make it structurally worse, because there is no driver who wants to clock out and stop the meter.

Particulate emissions are a real number that the all-electric story conveniently buries. Electric vehicles are heavier than comparable gasoline vehicles by roughly 20 to 30 percent because of the battery pack. Heavier vehicles shed more PM2.5, the ultra-fine particulate matter under 2.5 microns that lodges in lung tissue and crosses into the bloodstream, from tire wear and brake dust. Independent emissions research has consistently found that non-tailpipe particulate emissions from heavy electric fleets approach or exceed what a modern gasoline car puts out the exhaust. The fleet runs 24 hours. The dust settles on the same residential blocks where the cars stage.

Emergency response also degrades on a normal day. The Venetian Causeway incident already showed what a single stalled robotaxi does to a chokepoint with no weather involved. Outside the hurricane scenario above, every fire call, every ambulance run, every active police response now competes with a fleet whose default failure mode is to stop in the middle of the lane. The risk lands on every resident, not on the company whose software made the decision.

Enforcement is its own subsidy. Every Code Compliance hour spent chasing a staged Waymo is an hour not spent on illegal short-term rentals, broken streetlights, or actual resident complaints. The city is paying a public employee to police a corporate logistics decision, and the corporation pays only the ticket itself. The labor cost stays on the public ledger. Multiply that across a fleet that doubles every quarter and the math gets ugly fast.

Data extraction is the externality nobody is talking about yet. Every Waymo on a Miami Beach block is mapping it in high-resolution laser scans and multi-camera video continuously. Faces, plates, package deliveries, who walks the dog at what hour, which houses have which cars in the driveway. That data has commercial value. It is being collected on residential property without compensation and without consent from anyone outside the paid passenger. Florida has no current legal framework requiring Waymo to disclose what it retains, who it shares it with, or how long it keeps it.

Insurance socialization is the slow-burn one. When a self-driving car causes a collision, the liability picture is still contested. Fleet operators argue the software performed within spec. Drivers and pedestrians sue. Settlements get absorbed into broader auto insurance pools that every Florida driver pays into. Florida already carries among the highest auto premiums in the country. Adding fleet-scale autonomous risk to that pool spreads the cost across drivers who never use the service and never agreed to underwrite it.

Stacked together, the parking ticket starts to look like the price of admission for everything else. Alphabet pays $100,000 a year in fines and walks away with curb space, sleep, road capacity, air quality, emergency-response margin, enforcement labor, residential surveillance data, and insurance subsidy. The asymmetry is not subtle. It is the entire business model.

WHY MIAMI BEACH IS STRUCTURALLY WORSE

San Francisco at least has the Sunset, the Richmond, and the Bayview where deadhead vehicles can spread out. Miami Beach has none of that.

The island is roughly seven miles long and one mile wide at the widest point. Three causeways in and out. Once a Waymo crosses the MacArthur or Venetian and finishes a Lincoln Road drop, it has nowhere to go. Crossing back to a Miami staging lot eats deadhead miles and causeway capacity. So the fleet algorithm picks the cheapest option. Sit on a residential block in South Beach or Mid-Beach until the next ride hits.

Add the rest of the local context. Residential parking on Miami Beach was already broken before robotaxis got here. Tourist density, short-term rental conversions, and visitor permit programs absorbed every available curb. Code Compliance is stretched thin enforcing existing parking rules. The city does not have the bandwidth to write 589 tickets a year against a single corporate fleet on top of everything else.

Waymo’s cost is lower here than it was in San Francisco. The marginal harm to residents is higher.

WHAT THE COMMISSION CAN ACTUALLY DO

Miami Beach commissioners have a window right now while the fleet is still small. Three actions move the needle.

Require dedicated staging infrastructure. Waymo and its fleet partner Moove should be on the hook to lease commercial parking capacity proportional to fleet size. No staging on residential streets, period. San Francisco eventually got there. Get there first instead.

Geofence the residential zones. The city can require autonomous fleet operators to maintain digital exclusion zones on residential blocks during low-demand hours, enforceable through the fleet’s own telematics. Waymo already collects this data. Make compliance a permit condition.

Cap fleet size by neighborhood density. New York did this with rideshare drivers. The same logic applies to robotaxis with even more force, because there is no driver income argument to push back on. Fleet size is a corporate logistics decision, and the city has standing to negotiate it.

None of this is anti-technology. Robotaxis are a real safety improvement on impaired and distracted human drivers. The question is who absorbs the cost when the corporate logistics model leaks into public space.

THE WINDOW IS CLOSING

Waymo’s expansion is happening whether Miami Beach negotiates or not. The fleet on the island will grow. Highway access opened up two weeks ago, which means more cars crossing the causeways every day, which means more empty miles as each vehicle finishes its last paid ride of the night.

The cities that win this fight are the ones that move before the fleet hits scale. The ones that lose wake up at 3am to a Jaguar I-Pace humming on their block and realize the parking ticket is a feature.

Document what you see. Photograph staged Waymos on residential streets. Bring it to the next commission meeting. The window for getting this priced correctly closes once the fleet doubles.

There’s a reason. It’s the same reason the stock market keeps breaking records on artificial intelligence. The same reason every tech company on earth is racing to build data centers. The same reason your senator just gave a speech about the grid.

Most people are reading these as separate stories. They’re not. They’re one story, and you’re inside it whether you wanted to be or not.

By the end of this piece you’ll see the connection. You’ll know what’s actually squeezing the middle class. You’ll understand why the next decade is going to look different from the last fifty years. And you’ll know the one move still available to you before this gets harder.

The mortgage that saved you

When the housing market cracked in 2008, what made it ugly wasn’t the price drop. Prices fall in every recession. What made it catastrophic was the cascade. Adjustable-rate mortgages reset higher. Owners couldn’t pay. Banks foreclosed. Neighborhoods filled with for-sale signs. Prices fell more. More people went underwater. The cycle ran for two years until something broke.

That cascade can’t form today. Roughly six in ten American homeowners are sitting on mortgages locked at fixed rates below 4 percent. They’re not selling. They literally can’t afford to. Selling means buying again at today’s rates near 7 percent, which on the same house roughly doubles the monthly payment. So they stay. They eat at home. They cancel Netflix. They lower their standard of living in small ways. But they keep the house and they keep the rate.

That fixed mortgage is the wall holding up the American middle class. It’s why we haven’t had a 2008 redux even though everyone keeps predicting one. It’s also why every investor who bet on a housing crash over the last four years has been wrong.

But there’s a second monthly bill in your house. And it doesn’t have a rate lock.

The bill that doesn’t

For the last fifty years, the cost of electricity in this country has been almost flat once you adjust for inflation. Rates went up in dollar terms, sure. But after subtracting general inflation, real prices barely moved. Electricity sat in the same mental category as water. Boring, predictable, automatic.

That fifty-year run is ending. Not slowly. Quickly. And not because of any policy decision you can vote against.

It’s ending because of who just showed up at the meter next to yours.

Who showed up

The four biggest cloud companies in the world are Microsoft, Google, Amazon, and Meta. The industry calls them hyperscalers, because they run data centers at hyperscale. Between now and 2028, those four companies are spending almost three trillion dollars on data center buildout. Most of that money is committed. A lot of it is already in the ground.

Those data centers run on electricity. A staggering amount of it.

The federal agency that tracks national power statistics is forecasting the strongest four-year stretch of demand growth since 2000. In Texas, the state grid operator is planning for power demand to grow ten to fifteen percent every year. In the mid-Atlantic, which covers thirteen states from Illinois down to North Carolina, the planning number is three to five percent. For context, before this we considered one percent annual growth normal. We’re now planning for ten times that in the parts of the country where data centers cluster.

Here’s the part that matters for your bill. Utilities aren’t normal companies. They don’t compete on price. They’re regulated monopolies. A state commission decides what they’re allowed to charge. When a utility’s costs go up, they file paperwork with that commission. The commission approves a higher rate. The cost gets passed through to everyone who plugs into the wall.

That paperwork is being filed right now. In Virginia, Ohio, Texas, Georgia, and other states, commissions are approving rate hikes that explicitly cite data center load as the reason. Microsoft and Google aren’t paying for the new substations. You are.

Commercial customers feel this before residential customers do. Small and medium-sized businesses sit on a different tariff than your house, and that tariff moves first when the grid gets stressed. Their bills are climbing right now, this quarter. The federal inflation number won’t catch up for another six months. The bill changes before the headline does.

The buildout that can’t happen fast enough

If the grid could just expand to meet the demand, none of this would matter much. Hyperscalers would build data centers. Utilities would build out the grid. Prices would settle.

That’s not what’s happening. The buildout is hitting four hard limits at the same time.

The first is transformers. These are the metal boxes, some refrigerator sized, some house sized, that step electricity up and down between voltages so it can travel long distances and then safely power your home. The United States doesn’t make most of its big ones. We import them from Korea, Germany, and Japan. The wait time to order one is now two to three years. You cannot expand a grid faster than you can install transformers, and you cannot install transformers you don’t have.

The second is the line at the front door. When a new power plant or a new data center wants to plug into the grid, it has to apply, get studied, and wait its turn. The line is called the interconnection queue. In several major regions it is now five to seven years long. Approved projects are sitting there waiting for their slot.

The third is copper. Every wire, every transformer, every substation needs copper. Copper supply is tight. No major new mines come online for at least a decade.

The fourth is the people who actually do this work. Industrial electricians. Substation technicians. High-voltage linemen. The workforce is aging. The replacement pipeline isn’t there.

Stack those four constraints on top of each other and you get a buildout that physically cannot happen at the speed Wall Street is pricing in. Demand keeps climbing. Supply is rate-limited by reality. The thing that closes the gap is price.

Before you panic, let me draw a line

This is not a prediction of blackouts. The grid will keep working. Localized outages will happen, but the lights will stay on for most of you most of the time.

This is also not a hyperinflation call. The dollar is fine. The official inflation number will keep printing in a normal range.

What this is, is a relative shift. Electricity rises faster than wages. Faster than the official inflation number. Households without their own power source pay the difference. People with paid-off solar arrays don’t. The gap between those two groups widens, and it keeps widening.

The squeeze that every macro writer keeps describing is real. They’re just describing it from too far away. Up close, the squeeze is a line item on a piece of paper that arrives in your mailbox once a month.

Who feels it first

Some households are already protected. Homeowners with rooftop solar and battery storage have basically prepaid the next twenty years of their bill. Their variable cost for power is small or zero. The squeeze cannot reach them.

Everybody else is exposed. Renters can’t put solar on a roof they don’t own. Small businesses sit on commercial rates that move first. Veterans on fixed disability see annual cost-of-living adjustments that always run behind the actual cost of living. Retirees on Social Security same problem. Working families in apartments with electric baseboard heat in cold states are already getting hammered.

This is the population the macro writers gesture at when they say “the middle gets squeezed.” The conduit they don’t name is the meter on the side of your house.

The piece nobody is writing

Here’s where I’ll wander into territory most investment newsletters won’t touch.

Whoever scales their power grid fastest wins the artificial intelligence race. There isn’t a second answer. AI runs on electricity. A country that can’t deliver electrons to its data centers can’t deploy AI at scale, no matter how good its chips and its software are.

China understands this. China makes its own large transformers. It controls a major share of the world’s copper processing. It owns most of the rare earth supply chain that goes into wind turbines and grid hardware. China is electrifying at a pace the United States cannot match, because China has vertical control of its inputs and a government willing to permit, build, and energize on a five-year horizon.

The United States permits new transmission lines on a fifteen-year horizon if it permits them at all. We import our heaviest grid equipment from three suppliers in two allied countries. Our trades pipeline is broken. Our interconnection queues are stuffed. We’re running this race with one leg tied behind our back, and the leg is the power layer.

This isn’t just an investment story anymore. It’s a national security story. Energy security used to mean keeping oil tankers safe in the Strait of Hormuz. It now means whether your home grid can power the systems that run the economy.

What you can actually do

You can’t refinance the bill. But you can shrink your exposure to it.

The closest thing to a fixed-rate mortgage on power is rooftop solar combined with a battery. The math depends on where you live. In Texas, Florida, Arizona, and most of California, solar and storage pay themselves back in seven to ten years and run essentially free for fifteen more after that. In the Pacific Northwest or upstate New York the math is harder but often still works.

Don’t think of it as an environmental statement. Think of it as fixing your rate. The same buildout that’s driving your bill up makes the case for putting panels on your own roof stronger every quarter.

If you run a small business, look at your commercial bill and look hard. Most small business tariffs are flat rates that are about to get repriced. Shifting heavy loads like air conditioning, water heating, and electric vehicle charging to off-peak hours can cut a commercial bill by fifteen to thirty percent without spending a dollar on equipment. If you’re big enough to put solar and storage on your building, run the numbers this year, not next year.

The investing angle, the one Wall Street is selling, is fine. Buy the cooling vendors. Buy the switchgear makers. Buy the transformer manufacturers. Buy the nuclear restart plays. But understand the order of operations. Defend your own household first. Hunt for alpha second. The investing trade is a hedge on the same exposure you already have through your bill, and you can hedge it more directly through your roof.

Closing

I opened by saying most people are reading two stories when there’s only one. The story is this. The same buildout that’s making Microsoft and Google more valuable every quarter is making your power bill the most exposed line in your household budget. The mortgage is the wall that’s holding. The bill is the wall that’s about to crack.

You can’t stop the buildout. You can lock in your own rate before the price clears.

Stay sharp.