A five-year, $5 million grant to HathiTrust will create responsible, equitable, and transparent services that leverage AI for the public good and scholarly discovery. Authors Alliance and Northeastern University Library are partners in the effort.

HathiTrust has been awarded a five-year, $5 million grant for Transparent Books: AI Infrastructure for Discovery and the Public Interest from the Mellon Foundation. Authors Alliance is a partner in the project alongside Northeastern University Library.

For us, this grant continues work we began with the Public Interest Corpus: opening a path for research libraries to lawfully make the books in their collections usable for computational research and discovery on terms that support the goals of research, scholarship and access to knowledge, rather than on terms that benefit commercial intermediaries and technology companies.

The Public Interest Corpus, which Authors Alliance co-led with Northeastern and carried out with Mellon Foundation support in 2024 and 2025, was a planning project that asked how libraries can make the books they have collected and preserved lawfully available for computational research in ways that serve authors, scholars, and the reading public, instead of leaving book-based AI development to a small number of companies working from datasets of uncertain and contested provenance. Through workshops in Boston, New York, and Oakland, dozens of interviews, and the guidance of an advisory board of librarians, researchers, authors, and technologists, the project produced a framework for a library-stewarded, noncommercial books corpus. The report and supporting materials are available at publicinterestcorpus.org.

This next phase puts the lessons from The Public Interest Corpus into practice, using HathiTrust’s collection of more than 19 million digitized volumes as its foundation. HathiTrust will serve as the central partner and steward of the corpus. Northeastern will develop the tools and infrastructure for search, discovery, and text and data mining. Authors Alliance will lead the project’s policy and convening work.

Authors Alliance’s core mission is to support authors who write to be read, and who want their books to contribute to research and learning rather than sit unused. Our role in Transparent Books is to help articulate the legal and normative principles for use of the corpus that will do just that, informed by convening the authors, researchers, and institutions the project aims to serve. Through this project, we want to lower barriers for noncommercial, academic AI researchers so they can build models and tools that serve the goal of advancing research and benefiting the public. We seek to do so in a way that prioritizes the interests of authors in seeing their works and their ideas discovered, attributed, and built upon.

Over the five years of the grant, the partners will:

• Enhance Book Discovery: Apply AI tools to the HathiTrust corpus to improve descriptive metadata (e.g., subjects, languages, summaries) and text quality, and to develop interactive search to facilitate deeper understanding of books, both as individual items and in aggregate.

• Provide Verified Resources: Offer legitimate, verified, and transparent sources of text and images from books at scale for academic and noncommercial use cases, for purposes such as text mining, computational analysis, and the creation of specialized language models.

• Advance Shared Principles: Develop principles for legal frameworks, research integrity, accountability, and environmental impact to serve as a shared foundation for decision-making that centers academic values and the public good.

• Create Sustainable Services: Develop financially sustainable services that support academic and noncommercial research.

Transparent Books will help ensure that the tools shaping how people read and research in the coming years are built on a foundation of values of authors who want to serve the public with their work. We are grateful to the Mellon Foundation for its continued support, and to our partners at HathiTrust and Northeastern University. We will share more as the work develops, and we welcome authors and institutions who want to take part. For more information, see this announcement from HathiTrust.

The Office of Management and Budget’s new proposed rules for the administration of federal grants have caused a storm.

Published on May 29 as the largest rewrite of federal grant rules since the Uniform Guidance was consolidated in 2013, the proposal drew tens of thousands of public comments before the July 13 deadline. We wrote in June about what the proposed rule would mean for scholarly publishing and open access specifically, and our post last week highlighted our comments to OMB on those issues.

This post steps back and addresses a more basic question: is any of this legal, and what happens when OMB is sued?

We anticipate that these regulations, if finalized in anything like their current form, will face many legal challenges. Some will likely challenge whether OMB has the power to issue rules like this at all. Others may accept that OMB has some general authority but argue that particular provisions exceed that authority. Whether and how soon we would get direct answers from the courts is unclear – for one, it’s not certain how plaintiffs would establish standing, and other now well-worn tactics (e.g., efforts to move cases to a more friendly court like the Court of Federal Claims). But assuming the courts do pass on the merits of these kinds of claims, the rationale behind these challenges lies in certain particulars of federal administrative law that few authors spend much time thinking about, but because litigation would likely have a significant impact on ongoing research, we thought a brief explanation of some of these issues would be helpful.

Does OMB have this authority in the first place?

Presently, the OMB Uniform Guidance is just that—guidance for how other federal agencies will spend money on grants, cooperative agreements, and financial assistance, rather than a regulation binding grant-recipients directly. OMB suggests policies, but each agency has to implement them before they become legally binding. One thing the proposed rule does is consolidate power in OMB by stating that a new rule by OMB automatically becomes the rule for each of the agencies. So, when thinking about legal challenges, a good place to start is to ask whether OMB actually has the authority to dictate what rules other federal agencies impose on the grantees funded through their grant programs. This is a more complicated question than it may seem.

Agencies are creatures of statute; they have only the powers Congress gives them, and federal law requires agencies engaging in rulemaking to identify the statutory authority on which they rely. For this rulemaking, OMB points primarily to 31 U.S.C. § 503(a)(2), which directs OMB to “provide overall direction and leadership to the executive branch on financial management matters by establishing financial management policies and requirements,” along with a handful of related provisions, including 31 U.S.C. § 6307, which authorizes OMB to issue “supplementary interpretative guidelines” for grants and cooperative agreements.

Section 503 originated in the Chief Financial Officers Act of 1990, a statute aimed at fixing unreliable financial reporting and weak internal controls across the government. One would think that “financial management policies and requirements” in that context most naturally means things like accounting standards, audit procedures, and reporting systems, not substantive conditions on what grant-funded work may say or do. And “supplementary interpretative guidelines” in § 6307 is an odd phrase to hang binding, government-wide regulation on.

In fact, courts have recently had occasion to consider OMB’s authority in adjacent contexts, and the early results are not encouraging for OMB. In National Council of Nonprofits v. OMB, the plaintiffs challenged OMB’s January 2025 memo freezing nearly all federal financial assistance. The district court concluded that, to the extent OMB relies on § 503 for its authority to do so, that statute “strongly suggests that OMB occupies an oversight role” rather than one of direct command over agency spending decisions. A California district court quoted and followed that reading later in AFSCME v. OMB, No. 3:25-cv-08302 (N.D. Cal. Oct. 28, 2025). And in Woonasquatucket River Watershed Council v. Department of Agriculture, the District of Rhode Island enjoined an OMB-directed freeze of Inflation Reduction Act and infrastructure funds, finding that federal law did not authorize the “broad powers” OMB and the agencies asserted. Those cases are on appeal, but they show how courts are likely to approach the question: OMB coordinates and oversees; it does not, on its own authority, regulate grant recipients.

On top of this, the Supreme Court’s “major questions doctrine” means a rule of this scale will get an especially hard look. Under that doctrine, courts expect Congress to speak clearly before an agency may decide questions of vast economic and political significance. As the National Council of Nonprofits court put it, “[w]hen an agency claims to discover in a long-extant statute an unheralded power to regulate ‘a significant portion of the American economy,’” courts “typically greet its announcement with a measure of skepticism.” Util. Air Regul. Grp. v. EPA, 573 U.S. 302, 324 (2014) (quoting FDA v. Brown & Williamson Tobacco Corp., 529 U.S. 120, 159 (2000)). It is difficult to imagine a better candidate for that skepticism than a rule that converts decades-old “guidance” into binding regulation governing hundreds of billions of dollars in annual federal spending, on the strength of a 35-year-old statute no prior administration read to confer such power.

So will this question about OMB’s authority be the undoing of the proposed regulation? Here is where things get complicated, because the proposed rule isn’t coming just from OMB. The proposed rule is styled as a joint rulemaking: roughly forty grantmaking agencies, from HHS to NSF to the Peace Corps, are simultaneously proposing conforming changes to their own regulations. So, should the rules be challenged, it seems easy enough to say “this isn’t just OMB speaking, but all of the agencies, jointly.”

But as one legal analysis points out, the proposal says essentially nothing about what authority those other agencies are invoking (which they are required to do under the Administrative Procedure Act). More importantly, the joint-rulemaking structure would only address the current rulemaking. The whole point of reclassifying the guidance as an OMB regulation is that future amendments would take effect government-wide the moment OMB finalizes them, without any agency-by-agency adoption (though presumably, agencies would still coordinate under established review processes, e.g., under Executive Order 12866). To the extent that it is done so via authority granted by current agency heads, that raises all sorts of complicated questions about the extent to which they can delegate authority that Congress granted them to another executive branch official (in this case, OMB). The answers here are not clear.

Even with authority, the rule collides with other law

Even if a court accepts that OMB (or the agencies collectively) can legally issue this type of regulation, agencies still can only regulate in ways that don’t conflict with other statutes passed by Congress (and, of course, the Constitution). There are quite a few ways the proposed rules may seriously run afoul of the First Amendment’s prohibition on inhibiting free expression. These would include pretty explicit calls for agencies to engage in viewpoint discrimination on issues related to racial diversity, gender, and sex.

Going into all the First Amendment issues probably deserves its own post, so for now I’ll focus on one rather straightforward statutory conflict: the proposed rule on indirect costs awarded to grantees, an issue that the current administration has already been sued on (and lost) and that Congress has specifically spoken to.

In 2024, Congress passed the following:

“In making Federal financial assistance, the provisions relating to indirect costs in part 75 of title 45, Code of Federal Regulations, including with respect to the approval of deviations from negotiated rates, shall continue to apply to the National Institutes of Health to the same extent and in the same manner as such provisions were applied in the third quarter of fiscal year 2017. None of the funds appropriated in this or prior Acts or otherwise made available to the Department of Health and Human Services or to any department or agency may be used to develop or implement a modified approach to such provisions, or to intentionally or substantially expand the fiscal effect of the approval of such deviations from negotiated rates beyond the proportional effect of such approvals in such quarter.

Further Consolidated Appropriations Act, 2024, Pub. L. No. 118-47,§ 224, 138 Stat. 460, 677.

In Massachusetts v. NIH, decided January 5, 2026, the First Circuit affirmed a permanent injunction against NIH’s attempt to impose a flat 15% cap on reimbursement of facilities and administrative costs (the “indirect costs” that support research infrastructure). The court held that the cap violated the text above, concluding that “Congress went to great lengths to ensure that NIH could not displace negotiated indirect cost reimbursement rates with a uniform rate,” as well as HHS’s own regulations governing deviations from negotiated rates.

The proposed rule goes to some lengths to assert that it is not changing the indirect cost rate negotiation system, and states that it doesn’t want to hear comments on the subject. If that’s really true, at a minimum the proposed rules are ambiguous in some respects – e.g., where they propose major changes to what costs are allowed (such as subscriptions to academic journals) where those costs are ordinarily recouped through indirects. Lots of people (including ARL, AAU, and others) have expressed concern about this. We wrote about this question in June: the proposed revision to 2 C.F.R § 200.454, if applied to indirect costs would make subscriptions to “business, professional, academic, and technical periodicals” unallowable, which at major research universities runs to millions of dollars per year.

It would be odd to say that the government cannot change how indirect cost rates are calculated while allowing it to eliminate massive categories of the expenses that traditionally make up those rates. Congress directly responded to earlier attempts to slash indirect cost recovery. A court asked to review the subscription provision will have to decide whether accomplishing through the cost principles what Congress forbade through rate-setting is meaningfully different.

Conclusion

OMB has announced that it plans to make its new proposed rules effective on October 1 (an incredibly aggressive timeline, leaving barely any time for revision). Whether that actually happens, and what form the final rules take, will determine what actual legal challenges will look like. The comment record now before OMB is overwhelmingly negative, and perhaps OMB will change course, though I doubt it given that OMB Director Russ Vought has publicly and strongly defended key provisions. Congress may, of course, intervene. And the closer the implementation of the rules gets to the upcoming appropriation cycle, the more likely it becomes that Congress may act using the Congressional Review Act (a tool that allows Congress to review and override agency action). But assuming there is litigation, it will likely take some time to produce any meaningful clarification for researchers and their institutions. So for now at least, it seems the best we can do is plan for more uncertainty.

Judge Araceli Martínez-Olguín has closed the Bartz v. Anthropic case, issuing a 23-page order that grants final approval of the class action settlement, resolves several outstanding administrative details, and rules on attorneys’ fees. The fee question was the only part of the order to receive extended treatment.

As we noted in reporting on the May fairness hearing, the court gave each objector only a couple of minutes from the bench. The written order is more thorough, working through the objections at some length. As in the lead-up to the hearing, though, the most substantial objections came from people outside the settlement looking in. The court declined to entertain objections from non-class members, but it took care to remind those individuals that, precisely because the settlement does not cover them, they remain free to bring their own claims. That is already happening (e.g., the Chicken Soup for the Soul and Carreyrou suits are separate, individual actions) and the court’s order in approving the settlement may prompt others to follow.

Attorneys’ fees

The fee award for the plaintiff’s attorneys received extended discussion at the hearing and it was one of the main issues addressed in this final order. Class counsel initially sought $300 million, then later trimmed it to $187.5 million. In today’s order, the court approved an approximately $101.5 million payment to the attorneys. Measured against the original request, $100 million reads as restrained, even though it remains a very large fee by any ordinary measure. By my rough math, it works out to about $2,953 per hour across the 34,381.6 hours of attorney time the court credited. (In contrast, the court reduced the service award for each named author plaintiff from the $50,000 requested to $15,000—at the hourly rate quoted above, apparently valuing the class representatives’ own contribution to the case at something like five hours of their lawyers’ effort.)

Payments to class members

When class members will actually be paid is probably the most pressing question for many authors and publishers covered by the settlement. Under the Settlement Agreement, payments begin one business day after the “Effective Date,” defined as the date on which the time to file or notice an appeal of the Final Judgment expires. The Ninth Circuit generally allows 30 days to appeal. Assuming neither side appeals this order, that means funds should begin going out in a little over a month.

Subscribe now

This piece was written by our summer intern, Kalani Gaviola. We’re grateful for her contribution to the blog.

OpenAI publicly launched ChatGPT on November 30, 2022. Five days later, Sam Altman, CEO and co-founder of OpenAI, tweeted the following: “I am a stochastic parrot, and so r u”.

The tweet is a reference to a 2021 AI ethics paper, “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, which describes LLMs as stochastic parrots—able to convincingly repeat information based on probability, but unable to truly comprehend their outputs. While the paper ultimately uses this metaphor to frame its argument on the limitations and downsides of AI, Altman’s motives are more opaque. Mere days after the release of the most powerful AI chatbot to date, the words are eerie—are they a joke? A warning? A prophecy?

An informed reader might say all of the above.

I am a stochastic parrot, and so r u.

Indeed, in the years following the release of ChatGPT and the AI boom that followed, plagiarism has become one of the most pressing issues in academic spaces—and beyond—as writers use the ever-increasing capacities of AI to create in their stead, parroting AI outputs instead of truly processing and analyzing material. According to recent research from the College Board, college faculty estimate that nearly three quarters of students use AI to write papers or essays. Universities have adjusted plagiarism policies in the face of automated writing. In May 2026, for example, Berkeley Law enacted one of the strictest AI policies to date. This policy functionally bans use of AI in any part of completing course work aside from “identifying sources”.

In this scramble to combat academic misconduct, AI detection tools have become our best solution and de facto detectives. But how good are they, really? And how does the never-ending investigation of wrongdoing and ultimately, AI’s writerly abilities, affect students and authors alike?

In January 2023, Princeton student Edward Tian created GPTZero, a leading AI detection tool that he began coding mere weeks after the release of ChatGPT in 2022.

GPTZero “analyzes linguistic and statistical signals learned from human and AI text to estimate the likelihood that text was generated by an LLM”—a method used by various similar tools such as QuillBot, ZeroGPT, Originality AI, and more. As stochastic parrots, AI models function as prediction engines and tend to produce outputs with a statistical regularity human writings usually lack. For GPTZero, the signals to catch a stochastic parrot include perplexity (how predictable a sequence of words is), burstiness (variation in sentence length and structure), along with five other “indicators.”

Quickly, however, many pointed out the issues with this type of tool. While some markers may be indicative of AI use, large language models are ultimately trained on human writing and mimic it accordingly. In April 2023, several screenshots of the U.S. Constitution being flagged by GPTZero as “likely to be written entirely by AI” went viral, thus throwing the effectiveness of the entire enterprise into question. A few months later, a Stanford study showed that these tools are biased against non-native speakers, whose writing often scored higher in predictability and thus was flagged consistently as AI by detectors.

Even if one assumes accuracy, these tools are not necessarily effective. Gen Z is a tech-savvy generation. For every AI detector used by a teacher, there’s a tool that “humanizes” AI writing by purposely regenerating text to avoid AI markers, and often, you don’t have to look far. Even GPTZero, with the stated mission to “preserve what’s human,” recently joined Superhuman, the AI productivity suite behind tools like Grammarly, which allows users to “humanize” up to 1500 AI-generated words for free in mere seconds. The AI that enables detection can equally enable the generation of text specifically optimized to evade such detection.

Other methods are similarly unreliable. Authory claims to verify human writers by creating a custom style fingerprint that compares your writing to itself to avoid the pitfalls of generalization—but if all your available works were created using AI, the accuracy seems dubious. Some propose implementing AI text watermarks, which requires “embedding signals into generated text that are invisible to humans but algorithmically detectable from a short span of tokens.” But the watermarks are easy to remove. Other tools track the writing process by analyzing typing patterns, but, so long as you have some awareness of these tools, it seems easy enough to avoid pasting from AI and instead type it yourself while simply looking at what Claude generated on another tab.

Despite the unreliability of these tools, they have gained widespread popularity. In 2023, a UC Davis student’s story went viral after a professor reported him for cheating using AI. The professor’s accusation was corroborated by multiple detection tools including GPTZero, but was ultimately dropped after the student provided his document history and testified with the university’s honor court. The student claimed the experience caused him to have “full-blown panic attacks“, and if he had been less savvy in proving his innocence, he would have been subject to both failing the class and facing possible academic probation. A professor at Texas A&M similarly failed an entire class based on false AI use claims, causing the students to be denied their diplomas.

These types of stories have become commonplace—anecdotally, many students have been accused of using AI in some capacity. In academic spaces, the restrictions of AI seem reasonable; the purpose of education is to learn, and studies show that LLM users underperform at “neural, linguistic, and behavioral levels.” But some AI witchhunts have gone beyond academia, into fiction and creative spaces.

If you believe creative media’s goal is to entertain or make you feel something, it seems at least certain AI use should be encouraged. After all, AI’s narrative writing is now good enough that many are fooled by short passages from more advanced LLMs, even preferring them to human writing. Still, the internet crucifies the Duffer Brothers for allegedly using AI to write the famously underwhelming Stranger Things final season. A popular self-published novel, Shy Girl, is pulled by its publisher post-AI allegations. Despite the fact it’s impossible to know definitively whether or not AI was used in the creation of these pieces, the accusations continue. There are practical arguments against AI use in the creative space, whether that be via algorithms naturally creating less unique stories or LLMs stealing jobs from human artists, but the outrage seems to go beyond that.

Fanfiction is an artform that generates no money for the writers involved. It is known for being trope-filled and easy to read– you don’t have to be Margaret Atwood or Toni Morrison to have thousands of people read and enjoy your work. Still, many in the fan fiction community are ardently against AI use.

For example: in June 2026, an X (formerly Twitter) user called John Doe (@heatedrivalryai) created a thread informing netizens that they can search the HTML code of the popular fanfiction site Archive on Our Own for “font-claude-response-body” to identify if the author of a work pasted directly into the site from Claude. They also created a site skin that automatically turns the webpage red when a user clicks on a work that has this code embedded. Their post reached over 3M views within a week.

Though the method is shaky and does not in fact detect a mass of AI use (for example, if the author didn’t paste directly into the site from Claude, or simply used another LLM, it is ineffective), the fandom space was still taken by storm. The thousands of users in replies and quote retweets alike swarmed authors found guilty of AI usage. While some brought up the practical argument, condemning AI use generally for environmental reasons or citing fanfiction authorship as a valuable learning opportunity for young authors, many appealed to the intrinsic sanctity of human authorship.

“nobody is forcing ppl to write fics!!!” one wrote. “i don’t understand why ppl need ai to do it like pls just write fic from the heart.”

These tools and their ineffectiveness will surely continue to fester in our culture, but the extent of the outrage is interesting. It seems our discomfort with AI goes beyond practicality, and perhaps it should.

To many, our writing and our art is a fingerprint of our humanity—a transcription of our thoughts, and thus our personhood. It is our emotion, however predictable and replicable it may be, and our connection, and while some don’t mind having AI girlfriends, most people would prefer our relationships stay human.

You shouldn’t stop caring about human writing. You can use AI in moderation, as an additive tool instead of a replacement, but you should prize your effort, your art, and your learning, because without it, what are you?

The same goes for the AI detection tools—if you use these tools, use them as one piece of evidence instead of a death sentence, and use them only if you understand their efficacy and limitations.

Sam Altman may be a stochastic parrot, but you don’t have to be.

As we wrote in May, Elsevier is suing Meta over use of copyrighted works for training Meta’s AI models. It has a set of claims very similar to those in Kadrey v. Meta, but with a different purported class that includes authors and publishers who own copyright in academic journal articles and other academic works.

Along with two small updates in this case, we thought that now is a good moment to highlight and revisit some of the challenges of massive, class-action suits such as this to resolve copyright disputes.

The two updates:

First, on June 26, Meta filed a motion and supporting memorandum of law to transfer the case to the Northern District of California. We think Meta makes a strong case for transfer, given the overlap with Kadrey and the related litigation already pending in that district.

Second, on May 29, attorneys for Dr. Kenneth Saladin, professor emeritus of biology at Georgia College & State University, requested leave to intervene in the suit, with a proposed complaint, “to protect the distinct interests of textbook authors whose copyrighted educational works are included within the putative class alleged in this action.” (Proposed Complaint at 1.). The court quickly rejected that effort, but indicated that the request could be renewed once the court got to the question of class certification. Rather than wait, Dr. Saladin and several other textbook authors filed their own separate class action suit (in the Northern District of California), proposing a class that would include all textbook authors with a book used by Meta to train its models.

The transfer motion is fundamentally a battle for control over the litigation, albeit a minor one. The plaintiffs made a strategic decision to file in the Southern District of New York. Beyond the obvious convenience of their home forum, this choice had the added benefit of raising Meta’s costs. Meta, seeking more control, lower costs, and a court it has already been navigating in the Kadrey litigation, clearly prefers its chances in California.

The request to intervene is similarly about control, this time in terms of which law firms shepherd these cases and plaintiffs, who shapes the litigation, and who stands to profit from these class actions. These cases are big business for class action plaintiff law firms and they will not be shy about intervening or filing additional lawsuits, perhaps with an eye toward consolidation with suits already in motion.

The textbook authors’ efforts to carve out their own control over the Meta AI litigation gives a window into the possible conflicts of interest between some publishers and many authors, a problem we have long been interested in. To pull just one example from the proposed complaint:

“In Bartz, one of the major educational publishers sent a mass email to its textbook authors during the class opt-out period, which asserted that in many instances the authors should claim entitlement to only the sales royalty rate (e.g., 10%) of the proceeds, while warning authors against seeking third-party assistance and instructing authors to agree swiftly to ‘avoid delays.’” (Proposed Complaint at 17).

We must wait to see whether the new suit will simply be combined with the existing litigation in the Northern District of California (Meta would have good arguments for doing so, to avoid duplicative litigation) or whether the class issues in the Elsevier suit will be resurrected. But the issues raised by the textbook authors about how their interests diverge from their publishers in the outcome of the litigation are significant, and are symptoms of a larger problem that ought to be addressed in current and future class actions—many of these issues were present in the Anthropic v. Bartz litigation and remain unresolved.

Compounding the class action problems the Bartz class action left unresolved: who gets to sue on behalf of whom?

As copyright and AI litigation continues to wind its way through the courts (at least 125 active cases as of this month), we’ve noted that plaintiffs essentially have over a hundred legal laboratories running, to test which litigation strategies are most effective and which may be dead ends.

The most significant development of the last few years of AI copyright litigation is that we’ve seen the class action mechanism used to multiply damage claims, thereby increasing leverage for plaintiffs and putting intense pressure on defendants to find ways to avoid massive damage awards, potentially in the billions of dollars. Prior to 2023, the number of copyright class actions was minuscule in comparison, with very few cases approaching the scale of the current batch of AI copyright class actions in terms of numbers of class members.

Because Bartz v. Anthropic led to such a large settlement for plaintiffs, it provides a pattern that many would-be plaintiffs will want to follow. It deserves equally close study from those of us who are skeptical of where all of this is headed, particularly when one small group of plaintiffs claims to be able to represent the interests of millions of others.

We were concerned enough with class certification in Bartz that we filed an amicus brief asking the Ninth Circuit to review. Our brief was filed on August 7, 2025; the parties announced their settlement less than a month later, and the questions we raised were never resolved.

At the time, we said:

“Had the district court conducted such an analysis, it would have uncovered decades of research, multiple bills in Congress, and numerous studies from the U.S. Copyright Office attempting to address the challenges of determining rights across a vast number of books. It would have revealed how the complexity of years of publishing contracts makes ownership determinations fraught and prone to conflict. It also would have shown that rightsholders have genuine and significant differences of opinion about the legality of AI and the specific uses Anthropic has engaged in.”

Our concerns are just as valid today. The textbook-author intervention we note above is symptomatic of precisely the differences of opinion that went unresolved in Bartz and that linger in many of the cases currently moving toward their own resolutions, often with class certification as a step along the way. In Bartz, we’ve seen small glimpses of what this means at the end of the day for authors, both from an under-inclusion and over-inclusion standpoint. For example, many authors complained that they were entirely excluded from the class due to reasons of expediency and administrative ease (e.g., because their books weren’t registered or do not have an ISBN) and were actually prevented from formally objecting because of this). Others highlighted problems caused by over-inclusion, (e.g., by including in one class rightsholders such as publishers who have an interest in minimizing the claims of authors).

Questions that need answers if authors interests’ are to be protected in class action copyright suits

Building on our concerns in Bartz and with a purported class of journal authors and publishers in mind, here are some of the questions we would ask any court to consider. We begin with a top level question that we have come to believe was a glaring omission in Bartz:

1. What audit and transparency mechanisms will be put in place that will permit authors (including authors excluded from the class, perhaps unfairly), interested in the fair distribution of settlement funds, to examine precisely how funds are distributed to claimants?

One of the key weaknesses of the Bartz settlement is that there is no public-facing transparency mechanism in place to permit us to understand precisely how the settlement funds are ultimately distributed. There will be no open books to audit, no independent third party to spot check or otherwise verify that the more than one billion dollars in settlement funds is being apportioned to actual rights holders (not merely groups or publishers representing themselves as rights holders).

Yes, there is a settlement administrator and yes, the court can monitor the settlement. But the only people with real oversight of the process also have an interest in closing the book on the whole affair as promptly as possible. For the plaintiffs’ attorneys, who should be fully aligned with the interests of the authors they represent, thorough and exacting management of the process only eats into their bottom line. These are significant problems that Bartz failed to address.

For the sake of a clean number, let’s say the number of claimed works in Bartz is 440,000. If administering and verifying the allocation of funds for one claim took a modest 30 minutes (quite low for works where rights are contested), that would be 13.2 million minutes - 220,000 hours - or more than a century of full-time work!

2. Will the court address the tough in-the-weeds-details of rights allocations that matter for appropriately dividing settlement funds among class members?

In the Bartz settlement, we saw high level default presumptions and 50-50 splits, with a “we’ll figure this out” wave of the hand from plaintiffs’ attorneys, that left a host of questions unanswered. Will this be the norm for other class actions, or will judges in future cases pay more attention to the actual details of how rights are allocated? How will the settlement treat articles published under a Creative Commons license? What happens when the publishing contract doesn’t contemplate a settlement related to use in AI training, as most would not? In a complex mix of possible rights holders, publishers, authors, authors’ heirs, and authors’ employers, who is entitled to participate in the settlement? And how are uncertainties and nuanced questions resolved?

In Bartz, neither the plaintiffs’ attorneys nor the defendant had much interest in slogging through these details. Plaintiffs would aim for maximum class size to maximize payout, and the defendants (especially after negotiating a fixed payout sum) would want to maximize the size of the class to get the broadest possible liability release. With these priorities, how the money is precisely distributed becomes more of an afterthought.

3. Will the courts apply copyright law’s normal demands on litigation, or will class actions get a special pass?

To bring a copyright lawsuit for a US work, one must have first registered it with the US Copyright Office. This is copyright litigation 101. In most adversarial proceedings, defendants will typically vigorously scrutinize the registration of a work, whether it is valid, and whether it applies to the work at issue. The details matter a great deal.

In class action settlements, if Bartz is any indication, these details may receive more assembly line-level scrutiny and less heightened, high stakes crucible-level dispute.

But the details really do matter. To focus on one example: Group registration of serials is governed by 37 C.F.R. § 202.4. The regulation lets a publisher register multiple issues of a periodical at once, with the registration covering each issue as a copyrightable compilation. As to the individual works inside each issue, the regulation is more limited; it covers the authorship in the individual articles, photographs, illustrations, or other contributions appearing within each issue, but only “if they were first published in those issues and if they are fully owned by the author/claimant when the application is filed.” For a suit like Elsevier v. Meta, whether the articles at issue are actually covered by the proffered registrations could become a big deal.

Academic publishing has for at least two decades involved a sequence of distributions that precede the journal issue. Authors post preprints to SSRN, arXiv, bioRxiv, SocArXiv, and discipline-specific repositories before peer review. After acceptance, they often deposit author-accepted manuscripts in institutional repositories under their green open access rights. These distributions are usually unrestricted, indexed by Google Scholar, and consistently treated by academic readers as the operative version of the article. In many cases, these activities may constitute “first publication.”

To put it mildly, this is a fact intensive inquiry that will not be resolved easily and cannot be left simply to representations from plaintiff publishers. Put another way, in a great many cases, journal registrations will simply not extend to the individual articles within them.

Working out who is actually in the class is therefore not a matter of consulting the registration database for the issue. It requires, for each article, identifying whether the article was previously published, comparing those dates to the registration record for the issue, and assessing whether and when the article (as distinct from the issue) was timely registered. That kind of article-level inquiry is exactly what class certification’s predominance and manageability requirements are designed to scrutinize.

Where this leaves us

The Elsevier suit, like the others queued up around and behind it, will present a temptation to follow the trail Bartz blazed: certify a class, negotiate a settlement, distribute the money, close the book. Every step in that sequence embeds the questions above, and these questions don’t get easier when the works at issue are journal articles with preprint histories, layered contracts, and a tangle of possible claimants.

So, for whom is Elsevier suing? Right now, nobody can say with confidence. Unless a court insists on real answers about class composition and real transparency about distribution, nobody ever will. Until then, “the class” remains a convenient abstraction: precise enough to settle a case, but far too vague to truly represent the people whose rights justify the lawsuit. We will be watching, and where we can, we will keep pressing courts to ask these questions before pushing things forward.

Have you ever heard this: If you use 10 seconds or less of a song in a YouTube video, it’s fair use. Or this: You can copy 10% of a book for a class, and that’s fair use. Rules of thumb like these are common, but are they actually true?

In truth, neither of these statements tracks either the text of the fair use statute or any of the Supreme Court’s opinions on fair use. Instead, they come from private rules and beliefs about what may or may not be acceptable under copyright’s fair use doctrine.

People may turn to heuristics to help make sense of confusing legal standards like fair use because they are quick and user friendly, and seemingly provide clarity about what uses are fair. They may offer some comfort in the face of numerous statutory sections, judicial decisions, and expert opinions on what constitutes fair use: Just follow these rules and you’ll be on the right side of copyright law. But relying on rules of thumb like these instead of your own analysis of why a use is fair can lead users astray, possibly toward infringement instead of away, and can narrow the uses allowed under the law—restricting fair use, chilling creative creation, and limiting the production of the types of works that copyright law wants to encourage. Reliance on rules of thumb could also prohibit us from advising others in their own fair use analysis.

Fair use’s flexibility

Fair use draws much of its power from its flexibility that allows it to apply to all kinds of uses that Congress never could have thought of when it passed our current copyright law in 1976. US copyright law includes several targeted exceptions to copyright infringement that are limited in application and scope to specific areas. The fair use doctrine, in contrast, can apply to a broad range of uses and technologies, and can even bolster the application of other copyright exceptions. It provides an essential function in copyright law by helping to ensure that copyright doesn’t prevent the type of creation it is supposed to foster. Instead of a list of permissible uses, the fair use analysis requires a balancing test, weighing four nonexclusive factors against each other to determine whether a use is fair or not. As such, a wide range of uses can potentially be fair, depending on context.

However, fair use’s flexibility also makes assessing whether a particular use is fair or not challenging if there is not a definitive court ruling on exactly those facts. There will always be arguments both for and against fair use in every situation, so pursuing certainty can lead to people avoiding reliance on fair use. If you are at all risk averse, how can you put your faith in something as undefined as fair use? As none other than the esteemed jurist Learned Hand once said: “the issue of fair use . . . is the most troublesome in the whole law of copyright.” (Dellar v. Samuel Goldwyn, Inc., 104 F. 2d 661)

To avoid legal uncertainty, people seek out ways to make the fair use analysis simpler. In this environment, several quasi-legal myths have emerged.

The Classroom Guidelines and CONTU: Sources of fair use myths

One of these myths - the so-called 10% rule - is a sorely outdated misconception that derives from government sponsored discussions and related conversations about the nature of fair use from around the time Congress passed our current copyright law and codified the fair use statute.

In 1974, Congress brought together representatives from publishers, academic institutions, legal experts, and the public to consider issues related to copying, fair use, and new technologies, as the Commission on New Technological Uses (CONTU). Among other things, CONTU’s final report to Congress offered recommendations about fair use as it relates to interlibrary loan and library photocopying.

Around the same time, another group of authors, publishers, and academics prepared the Agreement on Guidelines for Classroom Copying in Not-for-Profit Educational Institutions with Respect to Books and Periodical (“Classroom Guidelines”) which offered recommendations to Congress on fair use in the academic setting. Congress eventually included these guidelines as part of its committee report on the new copyright law. And they continue to be errantly circulated by the United States Copyright Office in Circular 21, which has been widely criticized, including in this article by Sara Benson.

Notably, Congress adopted neither the CONTU final report recommendations on interlibrary loan nor the Classroom Guidelines’ suggestions. If Congress chose not to add these suggestions to the law, this suggests that Congress did not believe they belong as part of the law. Nevertheless, people today often seem to treat both like legal proxies. So, when even the Copyright Office seems to support their legal weight, it’s not surprising to see academic institutions often repeat the 10% or one chapter suggestion as if it is a legal rule.

But they are just suggestions, and we shouldn’t treat them as anything more. Indeed, with 50 years of technological developments and changes in user behaviors, it’s questionable what value they retain at all today. ARL published an excellent white paper on the CONTU guidelines in 2020 by Meg Oakley, Laura Quilter, and Sara Benson, arguing, as we do here, that it is time to move beyond CONTU and fully embrace fair use.

Some copyright myths stem from institutional policies or practices

So much of how we use and interact with copyrighted works today is through licensed digital materials, and license terms govern our use of those materials. Often, these terms restrict uses in ways that would conflict with what a court may consider a fair use.

Variances in institutional risk tolerance can affect the way that libraries interpret publisher-imposed restrictions that prohibit certain uses of licensed materials; conservative interpretations of contractual provisions have the potential to limit uses even beyond their intent. This can have implications for individual and collaborative research, hinder education and harm learning outcomes, and altogether negatively impact the work that libraries and academic institutions conduct.

Uncertainty about what a license allows can give librarians pause. For instance, in the context of copyright and accessibility, some librarians are hesitant to “remediate” (i.e., render accessible) a copyrighted electronic text document if there is no explicit contractual permission to do so. But copyright law allows libraries to remediate works for people with disabilities under Section 121, the Chafee Amendment, and Section 107 (fair use).

Publisher-imposed restrictions can push risk-averse institutions toward practices that are not required by copyright law, as illustrated in examples around creating and sharing of materials for people with disabilities. For instance, the AccessText Network (ATN) included a requirement for disability services offices (DSOs) to seek publisher permission for each remediated file requested by a student, and to destroy accessible files after the student graduates. DSOs may perceive these requirements as derived from the publishers’ copyrights rather than as contractual obligations. Once adopted, these institutional policies can persist even after publishers change their policies.

Copyright warnings imposed by publishers can also lead librarians and DSOs to second-guess whether explicit permission in a license or contract is needed in order to remediate works that are not available in accessible formats. A participant in a focus group of York University graduate students conducted by ARL and CARL highlighted not just how these copyright warnings show up, but also his understanding of how illogical they are:

“Sharing books, the texts go out of the way to be like ‘do not show this to anybody[.]’ And I was just thinking, if I didn’t need these services, and I took a book from the library, I would theoretically be able to pass it to another human being for them to read it. So the energy behind saying ‘do not share’ is a little bit weird…” (Read the full focus group transcript in the ARL/CARL Task Force on Marrakesh Treaty Implementation Final Report).

DMCA technical restrictions and online fair uses

Technical restrictions that online services have implemented in response to the Digital Millenium Copyright Act’s (DMCA) safe harbor provisions often regulate the ways people use copyrighted works online in ways that do not necessarily track the fair use statute, just as we have seen with the Classroom Guidelines, CONTU report, and publisher policies.

In 1998, at the dawn of the user generated web, Congress passed the DMCA to help facilitate the use of websites and services that host user content by creating protections for them against copyright infringement claims. If these websites follow the requirements in 17 USC § 512, they can avoid copyright infringement damages based on the actions of their users. In response, many websites and services have implemented technical measures to review and filter content that users upload onto the services.

These technical measures and other restrictive use policies create an environment that regulates the use of copyrighted works online more than the law does. In his article Internet Safe Harbors and the Transformation of Copyright Law, Emory Law professor and frequent commentator on the intersection of copyright law and technology, Matthew Sag, writes:

the balance struck by the traditional levers of copyright policy increasingly risks irrelevance. To a significant degree, the balance of copyright law has been overshadowed online by the system of intermediary safe harbors enacted as part of a general modernization of copyright under the Digital Millennium Copyright Act (DMCA) in 1998. Fair use and other axioms of copyright law still nominally apply online; but in practice, the safe harbors and private agreements made in the shadow of those safe harbors “DMCAplus” agreements) are now far more important determinants of online behavior.

We see this clearly from the supposed “rule” that you can use ten seconds (or some other number) of a song on a YouTube video and be fair use compliant. That in no way matches what the law says. Courts regularly tell us that there are no bright lines for how much of a work you can use and still be protected by fair use. Even using an entire work may be fair, depending on how you use it. But ten seconds or less is about how much of a copyrighted song you can use before triggering YouTube’s content filters. This is troubling. Unless a license applies to the use, it means that many wonderful, new, creative, productive, and fair uses of creative works may never be created for fear of running afoul of a content filter with no basis in the law.

These legal proxies can chill speech and restrict fair uses and other creative activities they ever happen

These rules of thumb are attractive because they purport to provide simple solutions and clear answers for what may or may not be a fair use, or seem to serve as reassurance for people working in copyright who may not have a legal background. As Amanda Wakaruk and her colleagues have found in their research on copyright anxiety and legal chill in higher education, those working in higher education are more worried about copyright than those outside the sector. But Wakaruk also found that copyright concerns can change practices, with real implications for research and pedagogy. Respondents who claimed that copyright concerns either caused them to avoid projects or hampered their projects described not using preferred images, readings or other preferred content in classes or research due to concerns about rights, and restricting access to digitized content even though it was unclear whether that was required by law.

It’s easy to understand why people may treat these short cuts as binding law and apply them uncritically. If fair use is confusing even to lawyers and judges, it is at least as arcane to the average person. Rules of thumb are quick, easy to apply, and seem to offer assurances that your use is fair. Why bother doing something different? However, fair use resists simple solutions and bright lines, and courts often remind us that fair use is extremely context dependent and requires a case-by-case analysis every time.

Consider these questions

Klosek and Wolfson offers this set of questions that libraries might consider when confronted by a copyright rule:

• What is the source for this rule—e.g. statute, regulation, institutional policy, contract term, other?

• Is the source of the rule the ultimate authority? (Is there an opportunity to renegotiate or relitigate? Is there a circuit split?)

• What is the rationale for the rule?

• Have changes in technology and business practices affected the rule’s rationale or implementation?

• Does this rule prohibit, narrow, complicate, or otherwise restrict activities that are lawful under fair use or other user rights?

• Who has a stake in perpetuating strict enforcement of this rule?

Conclusion

We think of these heuristics like urban legends: They may be based on something real, but over time they have taken on a life of their own, and they can affect the ways people behave, even if they don’t really exist. And like Bigfoot, Nessie, or that haunted house on the corner, the only way to fight them is to push back against the stories and focus on the truth.

We need to return the fair use analysis to the law itself. Copyright law seeks to balance the interests of rightsholders with those of users, and no doctrine captures that balance better than fair use. The system starts to break down when that balance tips too far in one direction. Fair use is essential to copyright working as intended, and these short cuts seem often to weigh too heavily against fair use, throwing off copyright’s balance of interests. But it doesn’t have to be this way. If we try to rely on the fair use law itself and best practices that draw directly from and align with the law and not these possibly helpful but also deceptive rules of thumb, maybe we can return the copyright balance to where it should be.

It is crucial to get the word out about the problems these copyright urban legends can pose for teaching, research, scholarship, and anything else that relies on fair use. Indeed, we regularly hear people say things about copyright that are simply untrue, and unchallenged or uncorrected, they can eventually develop into copyright urban legends that can effectively limit our rights in ways the law itself would not. We presented on this topic at the Kraemer Copyright Conference this summer, and we look forward to continuing the discussion. If you have examples or questions about these or other myths about copyright law, please get in touch with us through email or LinkedIn!

Katherine Klosek | kklosek@arl.org | https://www.linkedin.com/in/kklosek/

Stephen Wolfson | smw@upenn.edu | https://www.linkedin.com/in/stephenwolfson/

Last month, the Office of Management and Budget (OMB) issued a notice seeking comment on a proposed new rule. The rule would make highly consequential changes to the Uniform Guidance for Federal Financial Assistance. At last count, OMB had received nearly than 350,000 comments, and more than 50,000 were available to read. Today we filed ours.

The Uniform Guidance has occupied Title 2 of the Code of Federal Regulations since 2013. It replaced a series of OMB circulars, each governing a different class of federal grants. For example, Circular A-110 governed grants in higher education; Circular A-122 governed non-profits, etc. (A complete list is archived, here). The Uniform Guidance was an attempt to bring consistency and clarity to a necessarily confusing process. It largely succeeded.

We have recently written about the Uniform Guidance in connection with one specific section, found at 2 C.F.R. § 200.315(b)—the so-called Federal Purpose License (sometimes called the Government Use License), which is the provision that grants the government a license to use certain copyrighted works that it funds through grants. The white papers in our Legal Pathways to Open Access project, for example, go into detail on the mechanics of, and the basis for, the Federal Purpose License.

The OMB’s notice of inquiry comes in at 108 pages, and proposes many, many changes to the Uniform Guidance. Much of the text of the notice is highly political, and the tone of the notice reflects the present Administration’s political priorities. For the factual basis underlying their critical perspectives on previous executive branch work, they heavily on opinion pieces, internal documents, and executive orders from the current president.

Our comments focus on two points raised in the notice.

Our first point is on an internal contradiction in the notice itself. Presently, federal grant funds can be used to fund subscriptions to journals and memberships in professional societies. The OMB wants to change the rule to make those costs unallowable. The OMB also asserts they aren’t addressing indirect costs—ongoing lawsuits and injunctions mean their ability to change indirects is very limited.

But in academic environments at least, indirect costs often go to journal subscriptions by institutional libraries. We are taking OMB at its word that it is not modifying indirect costs, and so we read their comment on indirect costs to mean that the new rule would not apply to library expenditures of indirect funds. We think that is the only way to read the the notice to be within their granted authority. But as we noted in our comment, the contradiction has created confusion within our community, so we want OMB to clarify that is what they mean.

On the second point, OMB would like to change the rules of grant termination, to allow the Administration discretion to terminate any grant—including grants already in progress—if the grants no longer meet the administration’s priorities. Furthermore, that discretion would extend even to the government’s obligation to reimburse costs already expended under the grant. This is a very worrisome move. Grant research doesn’t correspond to the presidential election cycle, but under the new rule no grant would be safe after Inauguration Day. We don’t think this is permitted by the law, but it also is a terrible idea. We are asking OMB to leave the present rule—and its much narrower termination rules—in place.

There were many things we could have chosen to take issue with. We chose to focus on two. We could have chosen many more—there are other proposed changes we think are likely to be struck down if enacted, and still others that betray a terrible understanding of the conditions that allow a research community to thrive.

It is worth noticing, though, that for all the worrying proposals they make, the OMB’s proposed restructuring of the Uniform Guidance does not touch the Federal Purpose License, nor does it touch the public access policies that resulted from the “Nelson Memo.” At least so far, all indications are that this administration is still on board with the basic principles of public access to publicly funded works. In the end, that is at least good news.

As I mentioned above, the proposed new rules are sweeping, both in their changes and in their consolidation of power in OMB. We are not convinced OMB has the authority to issue many of these rules, and we anticipate litigation along those lines will follow their implementation. We’ll post more about our underlying legal concerns in the near future.

In the meantime, we will be following the Administration’s daunting task of reading every single comment and giving reasoned responses to the concerns raised. The timeline of the notice suggests they are unlikely to change anything, but the record established in the commenting period may help to control some of the excesses of the new rules in the future.

Last month, the Eleventh Circuit delivered its opinion in the case of Great Bowery v. Consequence Sound. The ruling is not terribly (ahem) consequential for the copyright world. This is not a hot button issue, and the opinion does not establish any new rules—at this stage of the litigation, it is a relatively simple question of standing. But the ruling does provide a useful instruction on the nature of “exclusive licenses,” the meaning of which has become a subject of some confusion in the scholarly publishing world.

Standing is a constitutional requirement for all federal cases. Under the federal courts’ interpretation of the Article III “cases or controversies clause,” courts cannot hear a case unless the party bringing suit can allege they have suffered some kind of injury. In addition, the injury must be an “injury in fact,” rather than merely a trivial or hypothetical one. In copyright, this means you must be the owner of an exclusive right in a work, or the court cannot hear your case.

The question here was whether the plaintiff had such an exclusive right. A well-known photographer granted the plaintiff, Great Bowery, the “exclusive worldwide right to license, market, and promote” certain of her images. However, in the same paragraph, the photographer reserved certain exclusive rights for herself over those same photographs. As part of their defense, Consequence argued that the photographer’s act of reserving rights for herself necessarily meant she could not have transferred an exclusive right. After all, Consequence argued, it cannot be an exclusive right if someone else has an overlapping exclusive right. The District Court bought this argument holding that Great Bowery had no exclusive rights under the contract, and so had no standing to sue.

The Eleventh Circuit held, I think correctly, that the District Court’s argument was “not quite right.” The Court did not go so far as to say that Great Bowery did have standing; rather, the Court ruled that an important assumption on which the District Court relied was faulty, and so the question needed to be reconsidered in light of the correct assumption. The Court remanded the case back to the District Court, which will now reconsider the standing question.

•••

What caught my attention about this case is an implication it has for understanding a feature of nonexclusive licenses in the copyright system, found in 17 U.S.C. § 205(e). In a nutshell, if a copyright owner grants a signed, written nonexclusive license, and then transfers the copyright in the work to someone else, the prior nonexclusive license is not cancelled out. As we’ve often written, Sec. 205(e) is the mechanism that makes the federal public access policy and many university open access policies tick. The idea is the author grants the institution a nonexclusive license in advance, so that if a publication agreement causes the copyright ownership to changes hands, the institution can continue to provide public access to the work.

Publication agreements sometimes contain wording where the grantor (normally the author) conveys an “exclusive license” to the publisher. An assumption I have sometimes seen is that one cannot assign an “exclusive license” to a publisher when there is a pre-existing nonexclusive license that overlaps with it. An exclusive license, the argument goes, cannot be exclusive if someone else also has rights that conflict with the grantee’s rights. But this reading is incorrect, and the Great Bowery opinion helps explain why.

The confusion arises from distinct, but related meanings of the term, “exclusive.” Under one meaning of the term, “exclusive” refers to something that is undivided. In that context, the party on the receiving end is assured there is no one with existing rights. In journalism, for example, an “exclusive story” is one given to a single news source. An exclusive story would be of little use to a newspaper if their competitor already had standing rights over the same story, so “exclusive” must take this first meaning to make sense. By contrast, in the context of copyright, an “exclusive right” is closer to its root word—it is the power to exclude any future uses, but can coexist with prior nonexclusive rights.

The term, “exclusive license” is not a term that shows up much in the law. Title 17 often refers to nonexclusive licenses, but generally avoids the term “exclusive license.” Under the copyright law, an exclusive license is by definition a “transfer of ownership.” There is no difference between the two terms—they are two ways of expressing the same thing. (Outside a single use in the definition section, “exclusive license” only ever appears in Sec. 114, and that section is a bit of a mess). An exclusive license, then, is a transfer of ownership of an exclusive right. And if you own an exclusive right under Secs. 106 or 106A, then you have standing to sue for infringement of that right. However, at least in copyright, there is no separate “exclusive right to sue”—one must actually own an exclusive right in order to be able to sue over its infringement.

Importantly for this case, one need not own the entire scope of an exclusive right in order to be a copyright owner. The law allows subdivision of exclusive rights. One party might own the exclusive right to reproduce a work in print, and another the exclusive right to reproduce the work in a digital format. For better or worse, the ability to subdivide exclusive rights is limited only by the parties’ imagination. As an example, in this case the Eleventh Circuit quoted the Nimmer treatise for the proposition that an “exclusive license” could grant a newspaper dealer the sole right (i.e., an “exclusive right”) to distribute a particular edition of a newspaper on a particular day on a particular corner, and that would allow the dealer to exclude anyone else from selling that edition at that time in that place.

The Court explained that, since the exclusive rights are infinitely divisible, the photographer could provide exclusive rights to Great Bowery, and at the same time retain exclusive rights for herself, simply by carving the rights with a finer blade. Put another way, the photographer could (and did) give Great Bowery exclusive rights, but with holes in those rights where she didn’t. She sold the bag of marbles, but kept the red ones for herself.

Though the case is about conflicting exclusive rights, the same logic applies when an exclusive right overlaps with a prior nonexclusive right. When the owner of, say, a research paper assigns an exclusive license in a work to a publisher, any prior nonexclusive licenses are carved out of the exclusive license by operation of law. That does not make the exclusive license any less exclusive. The license is still a transfer of copyright ownership, the publisher still receives an exclusive right, and the publisher can exclude any future uses of that right. But as Great Bowery illustrates, Sec. 205(e) carves out the narrow piece of intellectual spacetime that any prior nonexclusive licenses inhabit.

•••

Analogies to real property are sometimes problematic in copyright, but they are useful here. The ways a land owner can divide her land are somewhat more limited than they are for copyright, but in principle, the owner of a parcel of land can carve up ownership rights to that land in much the same way the owner of a copyright can. A landowner can divide her lot in two, divesting ownership of one plot while retaining sole ownership of the second. In much the same way, in Great Bowery, the photographer transferred certain exclusive rights to her photographs while retaining different exclusive rights for herself.

In this simile, a nonexclusive license more closely resembles an easement on the property, such as one allowing utility access to sewers or electrical conduits. Easements that “run with the land” remain in place even after the property is sold, but the fact that land is burdened by an easement does not make it any less the landowner’s property—the landowner can still exclude anyone else from using that land. Sec. 205(e) causes signed, written nonexclusive licenses to run with the land, as it were. And as with the land easement, a prior nonexclusive license granted for a work does not make the rights transferred in an exclusive license any less of a transfer of ownership—it just narrows the scope of the rights transferred.

It should be said that the infinite divisibility of copyrights creates many problems, as does infinite divisibility of land. In practice, land is not infinitely divisible—over several centuries, courts have narrowed the means for transferring land. Too much divisibility in land leads to problematic checkerboarding and fractionation, and too much divisibility in exclusive rights makes copyright, as Mike Godwin put it, “fractally complex.” I would love to see Congress fix this, but in the meantime the divisibility does at least help make sense of the nature of exclusive licenses.

•••

Looking forward, the case presents interesting questions about the nature of copyright licensing agreements as well as third-party rights to bring infringement actions on behalf of another. The District Court will undoubtedly consider them now that this standing question has been resolved. For now, the Eleventh Circuit has provided some useful guidance on some confusing terminology.

Grant administration rules are not where most authors look for news that affects their work. But, the so-called “Uniform Guidance” published in Title 2 of the CFR has a big impact on how universities account for federal money, from what counts as an allowable cost to how indirect rates are negotiated. The rules about what a grant can pay for, though, are also rules about how publicly funded research gets published and read.

On Friday, the Office of Management and Budget proposed revisions that would make sweeping changes to how and for what purposes grants can be used.

Much of it is overtly partisan, and I’m not commenting on that here. What I’d like to draw your attention to instead are a few provisions that would have a direct impact on how authors’ research is published and made available to readers.

The first addresses article processing charges, the fee that funds a large share of open access publishing; the proposed rule flips the default and treats APCs as unallowable except in narrow circumstances. The second is the cost of academic journal subscriptions, which universities have long folded into the indirect cost pool that supports their libraries. The proposed text appears to make those subscriptions unallowable as well, though, as explained below, it is far from clear that OMB means to reach institutional library spending. A separate provision, the federal purpose license in 2 CFR 200.315(b), is the legal mechanism agencies rely on to deposit grant-funded articles in public access repositories. It is left unchanged, which is also meaningful, though certain language in the proposed rule makes it unclear.

How these line items are treated will shape the economics of scholarly communication more than their placement in a grants regulation suggests. Below is an attempt to state more clearly what these proposed changes would do (and what remains unclear). As you read these provisions, I’d encourage you to collect your thoughts and consider contributing them to OMB (comments are currently due July 13, 2026, though an extension has already been requested).

1) Looking first at the structure of the proposed rule, OMB seems to be taking a far more assertive role with respect to grant administration than it has in the past—it seems to be transforming the uniform “guidance” into uniform “rules.” Part B.2. of the proposed rule talks about clarifying the regulatory structure (I admit the existing structure is confusing!), by stating (a) that Title 2 of the CFR is not merely “guidance” (as it presently clearly states) but is in fact a regulation that must be followed by other agencies, and (b) that changes to Title 2 will promulgate automatically, not waiting for agencies to adopt them themselves. It includes a lengthy explanation of its authority to do this (e.g., “Congress authorized OMB at 31 U.S.C. 503 to set government-wide requirements for grants administration, and agencies must follow the OMB requirements in their award programs.” (emphasis mine)).

Whether OMB actually has this authority to directly issue its guidelines as binding regulations may end up as a matter for the courts. Existing litigation around indirect cost rate caps and funding freezes set by agencies has only obliquely touched on it. ¹

Who knows how this will play out, but the result seems to be less agency variation (though the tail end of the rulemaking makes clear there will still be some), and a more directive approach from OMB.

2) On APCs, the default rule is essentially flipped. Presently, grant administrators are permitted to charge Article Publication Charges from the grant funds; the proposed rule would make APCs (or “similar fees such as open access fees for professional journal publications and other peer-reviewed publications”) unallowable under Federal awards, except in two cases:

(1) when a federal law requires otherwise, or

(2) when they are approved by the agency on a case-by-case basis.

Aside from APCs, the other allowed publication and printing costs (described in 2 C.F.R. § 200.461) are essentially left alone. Presently, publication costs are permitted when (1) they report on work supported by the Federal government; and (2) the charges are levied impartially; and awardees may charge costs at closeout for costs incurred outside the period of performance. The proposed rule keeps these rules, but restates them.

It seems to me the most important question under the new proposed rule is what does “case-by-case” mean? If it simply means that, where the particular agency decides to allow APCs in the first place, APCs as a category must be budgeted for and approved in the grant, then this doesn’t seem like a monumental shift for APCs paid by many research projects. If, however, it means something more onerous—e.g., prior approval on a publication-by-publication basis—then the rule would be so difficult to comply with as to effectively mean a prohibition on APC payments in all but the most extraordinary cases.

Whatever “case by case” means, it seems to me that agencies would retain the ability to fund OA publications that are specifically considered for OA, e.g., as NEH has done by funding specific books to make them open access.

But, it also seems that, combined with the change to Sec. 200.454 (below), this new text could make it difficult for universities to include article processing charges or similar as part of bigger transformative agreements in their indirect cost calculation since such agreements aren’t amendable to case-by-case assessment and because they are typically funded as indirect cost rather than direct costs.

One more thing on APCs: the proposed rule aims to limit the allowability of APCs strictly based on the relevance to the program that makes the grants. As the proposed rule explanatory text asserts, “In many cases, such activities are discretionary, vary widely in scope and costs, and may serve institutional, professional, or reputational interests rather than the specific objectives of the Federal program.” But just how direct the relationship between paid-for open access and the grant program’s objectives must be remains unclear.

3) Spending on academic journal subscriptions is restricted (but we don’t know to what extent)

Presently, 2 CFR § 200.454 reads: “Costs of the recipient’s or subrecipient’s subscriptions to business, professional, and technical periodicals are allowable.” The proposed new rule would make those costs unallowable, and would add academic periodicals as unallowable costs.

Read in isolation, this proposed rule change would represent a dramatic change in the types of costs that universities can charge against grants for subscriptions. For most universities, subscriptions to academic journal costs are largely included in the indirect cost pool, via amounts representative of library expenditures. This spending can be quite large – most major research libraries now spend the bulk of their collections budgets on electronic subscriptions, totaling in many cases millions of dollars per year.

If the rule means subscriptions could not be used as part of the university’s indirect cost base, this would represent a major reworking of indirects for many universities (For more on the subject, we’ve written about indirects before, such as in this post last year)

However, it’s not entirely clear that this is meant to apply to library subscriptions, as opposed to one-off subscriptions just for particular research projects. This is for a few reasons.

First is the context in which it appears. This is a section that also disallows costs of things like country club and dining club memberships, all of which are the kinds of things providing narrow, individual benefits, while still specifically allowing memberships in professional, technical, and civic or community organizations—not the kinds of things that are used by the entire institution. The explanatory text for this section is spartan. OMB merely notes that “Under the proposal, all other costs, including the costs of subscriptions or memberships in country clubs or organizations whose primary purpose is lobbying or issue advocacy, are unallowable.” It’s not impossible that OMB means to also limit broader institutional subscriptions here, but the placement would be strange.

Second, in the broader explanatory text OMB includes about the overall rulemaking, OMB include a lengthy discussion of the existing review of indirect cost rates, noting an extensive process for reviewing and changing indirect rates. This includes, OMB notes, legislation in the 2026 appropriations bills to essentially keep negotiated indirect rates static for now (OMB doesn’t mention all the recent litigation about indirect costs, but surely that is also part of the backdrop). Given the existing process, OMB states that the proposed rule does not change the negotiated indirect cost rate system, and that OMB will not consider comments on that subject.

Taking OMB at its word, it would seem odd then to include, buried in this section, a prohibition on allowable journal subscription costs, ordinarily paid for through indirect cost rate pools, that would upend many universities’ negotiated indirect cost rates. If the intent or effect is to make such a large category of costs unallowable, that seems to me the kind of change that would be immediately challengeable in court as a violation of Congress’s direct instruction not to modify indirect cost rates in the most recent appropriations bills.

4) The federal purpose license in 2 CFR 200.315(b) is unchanged.

This is important because it is the primary legal provision by which federal agencies obtain the necessary licenses from authors and universities to publish research articles, allowing them to share results freely via public-access repositories. We’ve written extensively about how this license system works and what its limits are here.

We do observe that the new proposed rule would update 2 CFR 2900.13, which references the CFR section where the federal purpose license is located (sort of; see below). The new proposed text provides that “In addition to the regulation set forth in 2 CFR 200.315(d),” (emphasis mine) the Department of Labor (and only the DOL) requires intellectual property produced under discretionary awards to be openly licensed.

However, this isn’t a meaningful change from the Department of Labor.. The update is merely to conform to the overall new structure OMB is proposing, changing “guidance” in the existing version to “regulation” in this new version. As Eric Harbeson explains in his paper “Open Access and Federal Information Policy” (p. 20) DOL has had this exact same open licensing language in place for some time, at least since 2021, and the Dept of Ed has similar open licensing requirements.²

1. Compare Ass’n Of American Univ. v. Nat’l Science Found., 788 F. Supp. 3d 106, 131 (D. Mass 2025) (“to the extent Plaintiffs argue further that NSF is statutorily barred from imposing any across-the-board cap on indirect cost rates, the court declines to resolve the issue where the parties have not addressed whether 31 U.S.C. § 503 vests exclusive authority in OMB to establish policies regarding reimbursement of indirect research costs in federal grant awards,”) with National Council of Nonprofits v. Office Of Mgmt., 775 F. Supp. 3d 100, 126 (D. Mass. 2025). (“DOD’s authority to issue grants is limited by OMB regulations governing the administration of those grants. See 10 U.S.C. § 4001(b)(1) (permitting issuance of grants “in accordance with chapter 63 of title 31,” 31 U.S.C. § 6301 et seq.); see also 31 U.S.C. § 503(a) (granting OMB authority to “establish governmentwide financial management policies for executive agencies”)). In cases involving a blanket funding freezes, the courts have interpreted this provision to say that OMB has an oversight role, but not more. See National Council Of Nonprofits v. Office Of Mgmt., 775 F. Supp. 3d 100, 126 (D.D.C. 2025) (“the structure and provisions of Section 503 strongly suggest that OMB occupies an oversight role. Defendants have not pointed to specific authority that allows it to unilaterally pull the plug on nearly all federal monetary flows.”). ↩︎

2. For those really into the details, note this minor oddity: the citation to § 200.315(d) rather than to the whole of § 200.315 is perplexing. Section 200.315 as a whole addresses rights in intangible property. Section 200.315(b) contains the “Federal Purpose License” that supports public access to scholarship. Section 200.315(d), however, applies only to the reuse of data. It provides that “The Federal Government has the right to:(1) Obtain, reproduce, publish, or otherwise use the data produced under a Federal award; and (2) Authorize others to receive, reproduce, publish, or otherwise use the data for Federal purposes.” As Eric notes in his 2024 article: “It is possible this is intended to supplement only the provision of Sec. 200.315(d), which applies to data supporting works, not to final products; however, by its plain language Sec. 2900.13 applies to all intellectual property produced under DOL grants, so it is most sensible to interpret the section as applying as a supplement to 200.315(b).”
   ↩︎

“We say that a model has “memorized” a piece of training data when (1) it is possible to reconstruct from the model (2) a (near-)exact copy of (3) a substantial portion of (4) that specific piece of training data. We distinguish memorization from “extraction” (in which a user intentionally causes a model to generate a (near-)exact copy), from “regurgitation” (in which a model generates a (near-)exact copy, regardless of the user’s intentions), and from “reconstruction” (in which the (near-)exact copy can be obtained from the model by any means, not necessarily the ordinary generation process).” A. Feder Cooper and James Grimmelmann, The Files are in the Computer: On Copyright, Memorization, and Generative AI (2025)

The above quote is from a fairly recent essay on the topic of Generative AI and memorization. It’s worth a thoughtful read, as are many of the articles written by or in collaboration with Cooper and Grimmelmann (see also Talkin’ ‘Bout AI Generation). They go into far more detail and depth on the topic of memorization than I can, though hopefully what I write here will be a useful entrypoint for some readers. As they note in their introduction, they are not taking positions on this technology and instead seek to describe it with accuracy (“We seek clarity, precision, and technical accuracy”, Cooper and Grimmelmann at 147).

Unlike them, I plan on taking some positions. Memorization is having a moment in the news, with grand pronouncements about its implications for the growing number of AI copyright suits: “Large language models don’t “learn”—they copy. And that could change everything for the tech industry.”

While Cooper and Grimmelmann are making every effort to not take a position on memorization, it is certain that memorization is already playing a meaningful role in current AI litigation, will likely inform future litigation related to outputs (e.g., substantially similar infringing outputs or infringement based on memorized inputs), and is being raised as a specter to question whether the technology itself can be a fair use. In this post, I am most interested in taking on the question of whether memorization forecloses the possibility of this technology as a fair use.

Memorization as inherent vice

To be absolutely clear, memorization happens in all models. As Cooper and Grimmelmann put it: “all generative-AI models memorize some portion of their training data.” (Cooper and Grimmelmann at 157). As a user of the models, especially the image generation models, it’s not hard to find examples of memorization. Below are three examples of Leonardo da Vinci’s Mona Lisa, one from Wikipedia, one generated by Google’s Gemini (2 prompts: “generate an image of da Vinci’s Mona Lisa”; “just the painting please”), and one generated by OpenAI’s ChatGPT (prompt: “generate an image of da Vinci’s Mona Lisa”). Without information to the contrary, I will presume that the generated images are products of memorization of Mona Lisa in the training data, and that this particular instance of memorization has been permitted by the AI companies, possibly facilitated and enhanced through training and not filtered at the output stage by guardrails due to the Mona Lisa’s public domain status.

(Note: I use the Mona Lisa because the demonstration is uncontroversial and clear, but memorization can be found with other, in-copyright works.)

Memorization is important to would-be plaintiffs because, it may be argued, memorization helps establish the elements of a copyright infringement claim. ”To establish infringement, two elements must be proven: (1) ownership of a valid copyright, and (2) copying of constituent elements of the work that are original.” (Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340, 361 (1991))

Memorization in AI models is also of particular interest to those suing AI companies if it can move the needle on the fair use analysis. Of particular interest is how it may affect the fourth factor fair use analysis, which focuses on the effect of the use on the potential market for or value of the original copyrighted work. For example, if a model would allow you to read or view the entirety of an in-copyright work, then it can be easier to argue that it is serving as a market substitute for that work, which is often a central consideration in any fair use claim.

As we’ve noted in previous posts, even in the absence of evidence that outputs are substantially similar to input texts used for training, we’ve seen the market dilution concern in Kadrey v. Meta, where Judge Chhabria wrote:

“As for the potentially winning argument—that Meta has copied their works to create a product that will likely flood the market with similar works, causing market dilution—the plaintiffs barely give this issue lip service, and they present no evidence about how the current or expected outputs from Meta’s models would dilute the market for their own works.” (Kadrey v. Meta opinion at 4). (This is a novel theory and one that has been deservedly questioned - see Ed Lee’s Copyright Dilution Under Constitutional Scrutiny)

To address the absence of evidence raised in Kadrey, we are now seeing moves towards surfacing potential market harms based on memorization. In Alignment Whack-a-Mole : Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Models, the article authors detail their efforts to extract text from frontier LLMs, using a finetuning approach that, according to their research, enhances their ability to extract blocks of text from books used to train the models. (“In the within-author setting—finetuning and testing on books by the same author—we find that finetuning unlocks latent memorization, enabling all three models to regurgitate massive amounts of verbatim text from held-out books, in some cases reproducing as much as 60% of an entire book. More alarming, this effect generalizes cross-author: training exclusively on Haruki Murakami’s books enables substantial extraction from over 30 other authors regardless of genre—in some cases reproducing over 80% of a book’s verbatim content, with single regurgitated stretches exceeding 460 words.”) (Alignment Whack-a-Mole at 3)

The article authors make the connection between their research and ongoing litigation explicit:

“There is another kind of market harm, not at issue in those cases, but which the present study may bring to the fore. A key factor in Bartz and Kadrey was the absence of evidence that the models trained on copied works generated outputs that reproduced the source works. But what if the outputs did reproduce the source works? What if users, with little effort, could extract substantial portions of the source works? The “regurgitations” are verbatim, or highly similar, copies that could well substitute for the source works. For example, why comply with a paywall, when one can prompt an AI system to deliver the content unencumbered by access or use restrictions? Would the AI developers’ failure to secure their systems against regurgitation-generating prompting undermine their defense on the fourth fair use factor?” (Alignment Whack-a-Mole at 11)

As a brief departure from this discussion of memorization, I’d like to also note that two of the authors of the Whack-a-Mole article, Tuhin Chakrabarty and Jane C. Ginsburg, also collaborated on another article, Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers, which looked at reader preferences and found statistically significant reader preferences for finetuned AI-generated works. Again, the authors concluded that their findings held significance for fourth factor fair use analysis.

There is clearly interest in memorization related fourth-factor evidence, and the pairing of Whack-a-Mole with Readers Prefer Outputs are two of the clearest examples of this I’ve been able to find. The animating force behind these inquiries, and the great consequence of memorization for these authors, is to make the case that AI models may fail to qualify as fair use because they offer impermissible competition with originals under the fourth fair use factor. The argument is that this happens through the profoundly negative impact AI has on the markets for and value of in copyright works, whether it is because the models risk reproducing too much of the in copyright works or because we ultimately prefer the outputs of the models.

[Below, I will offer a fuller counter to this view of the world. For those who prefer not to wait, I will offer a preview: memorization should be embraced rather than suppressed. The more, and more diverse, the material models retain, the more users can do with them. Paired with strong protections for users, this approach avoids two failures at once. AI neither overcorrects by foreclosing legitimate uses nor gets defined solely by commercial interests.]

Where do we draw our lines?

In both the Cooper and Grimmelmann essay and the Whack-a-Mole article, the authors nod to prior fair use litigation, which found that reproductions of limited portions of in-copyright works did not preclude fair use (see Authors Guild v. Google Inc., 804 F.3d 202 (2d Cir.), 2015). It’s hard to not think of Google’s “snippet views” when considering memorization and the portions of in-copyright works that the Whack-a-Mole authors point to as potentially winning evidence of market harm. While the Whack-a-Mole authors make efforts to show that the memorized portions of in copyright works potentially present a far more serious threat and that AI model users might attempt to recreate whole portions of in-copyright works, I am ultimately left feeling that this is the same rhetoric that we see each time a new technology promises large scale disruption.

Earlier in this post, I said that I would take some positions. Here are the positions I’d like to take at this time: (1) I don’t want to deny that AI models are disruptive and may ultimately do incredible harm; (2) I believe memorization will be a point of contention so long as the AI copyright lawsuits continue to wind their way through the legal system; (3) With some caveats, I think that overstating the threats from memorization is the wrong approach to contending with the potential harms (economic and otherwise) presented by AI and may ultimately be a strong disservice to the public interest.

Position 1

My first position is hard to argue with. It’s abundantly clear that virtually everyone is concerned about AI, its potential for monumental disruption, and the need to find solutions to its capabilities. OpenAI recently released Industrial Policy for the Intelligence Age: Ideas to Keep People First, which grapples with the social implications of AI and that these technologies could wreak enormous havoc on the economy and livelihoods of people everywhere (though in a way that one critic called “hazily sketched” and “reads a lot like something that ChatGPT would spit out, if you asked it to research ideas for combating AI-induced inequality for 10 minutes.”) Anthropic recently developed Claude Mythos and has limited its release due to substantial concerns around cybersecurity threats it could facilitate (See also Project Glasswing). These concerns, well documented, seem credible.

Below is a sampling of ongoing litigation related to AI. The technology is touching many facets of our lives; coupled with the rapidity and sometimes carelessness of its expansion, we are seeing a recipe for many types of harm.

I could go on with lists, databases, and articles and reports. The boom in AI over the past several years has been well documented, examined, and has created a simultaneous boom in commentary. If I tried to argue that AI is not likely to be disruptive and cannot, will not, cause damage, I think you’d have to extract my head from the sand.

Position 2

There are now roughly 100 copyright lawsuits against AI companies in the United States. That number will grow. Unless there are clear signs that it is not a winning strategy, we will see variations of memorization raised in more suits. Some litigation where memorization has already played a prominent role:

• If you look at the original complaint in The New York Times Company v. Microsoft, memorization is a key part of the factual allegations: “GPT-4 copied this content and can recite large portions of it verbatim…these examples represent a small fraction of Times Works whose expressive contents have been substantially encoded within the parameters of the GPT series of LLMs. Each of those LLMs thus embodies many unauthorized copies or derivatives of Times Works.” (Complaint at 31-32)

• In their motion for partial summary judgment, the plaintiffs in Concord Music Group v. Anthropic emphasized the role of memorization: “During training, Anthropic endeavored for its AI models to “memorize” Publishers’ lyrics so the models would output those lyrics. An internal Anthropic report recognized that AI models like Claude “memorize A LOT, like a LOT.””(Motion at 14).

• In Copyright’s Jagged Frontier, Matthew Sag details both lawsuits based on character memorization, such as Disney v. Midjourney (“It included dozens of clearly recognizable pictures of Princess Elsa, Bart Simpson, Darth Vader, The Mandalorian, Baby Yoda, Shrek, the Minions, and more.”) and other instances where models like OpenAI’s Sora2 produced images of characters like Snoopy and Pikachu (Jagged Frontier at 31).

For those of you interested in a deeper dive into this topic, Sag’s article is a worthwhile read. Not only does he provide a fuller picture of instances of memorization, he also helps to drive home a point that is made elsewhere: memorization is a product of model learning and we will often see the memorization of a block of text from a novel, or a beloved character, in the way we will see the useful and desirable byproducts of model training. “Memorization presents a stubborn technical challenge because most attempts to meaningfully curb the retention of copyrightable expression would also blunt a model’s capacity to learn a wide range of useful and unobjectionable material.” (Jagged Frontier at 29)

Given that the perils of memorization are intimately tied up with the benefits of AI models, it is hard to imagine that we will see instances of memorization go away completely. It seems far more likely that we’ll see swiss cheese AI models, holes where all of the most pernicious/litigation provoking (for users and the AI companies themselves) memorization has been excised or guardrailed unless licenses can be obtained. Tens of thousands of little line drawing exercises, ever evolving, likely fairly opaque to individual users, a writhing, pulsing, beautiful and ugly mess.

Position 3

At a high level, I want to first address those who suggest that memorization is likely to yield problems for defendants in the fourth factor fair use analysis. Returning briefly to Alignment Whack-a-Mole, the authors of the article suggest that, so long as memorization takes place in the models, it could weaken the fair use defense of those models (“As the court acknowledged, no matter how “transformative” the use, if its implementation depends on inadequately secured copies, the threat to the copyright owner’s market could offset the transformativeness.”) (Whack-a-Mole at 12) (internal citation omitted).¹

A regurgitated passage without the surrounding work isn’t competing in the same market as the book. The Authors Guild v. Google court understood this; the snippet view one sees in Google Books doesn’t substitute for the book. The reproduction of limited passages of books is also not a bar to fair use. If you spend enough time looking at digital collections and online archives, if you spend enough time thinking about copyright, you will see examples of fair use and entire in-copyright works everywhere. In Bill Graham Archives v. Dorling Kindersley Ltd, whole Grateful Dead posters were reproduced as historical artifacts within a cultural history book; the court found that this was a fair use. Authors Guild, Inc. v. HathiTrust - entire books used to create, among other fair uses, a searchable digital archive - again, a fair use. Kelly v. Arriba Soft Corp. - full images in thumbnail form - the foundation for image search that we rely on today - fair use. Ordinary, pro-social uses that have never been litigated such as (to pick one example) New York Philharmonic Digital Archives provides access to a range of works, some still in copyright and yet available in their entirety.

Of course, I know that we will have to draw lines and that harder cases are on the horizon. To spin up just one example inspired from Sag’s Jagged Frontier article, what happens when a user of one of the models tasks an agent with publishing a daily online comic strip, with millions of readers, featuring the travails of Detective Snoopy and his trusty sidekick, Pikachu?

Detective Snoopy and Pikachu comic generated using Google’s Gemini

Today, such scenarios are both easy to imagine and execute. In the future Sag realistically imagines, these scenarios will be addressed by some combination of guardrails and licensing regimes. Sometimes a guardrail will prevent a given use categorically. For the vast majority of uses, he sees a version of Content ID for AI, one where users have greater freedom of expression against a backdrop of negotiated licenses. “Licensing the rights to turn off targeted copyright safety measures and allow end users a freedom of expression that would otherwise cross over into copyright infringement has the potential to be a billion-dollar industry.” (Copyright’s Jagged Frontier at 10).

Sag’s account is probably correct as a prediction. The question is whether it should also be accepted as a prescription. Professor Katrina Geddes’s Engineering Semiotic Democracy is the strongest articulation I’ve found of an alternative view, not necessarily because it’s where things are heading, but because it helps show what can be lost if we let memorization-anxiety drive us to a licensing-mediated default.

AI is going to allow a lot of people who could not participate in cultural production to read, write, make images, and reach audiences in ways that were unfathomable previously. Geddes calls this phenomenon “Semiotic Democracy”—“When every member of a socio-political community has a roughly equal capacity to contribute to, participate in, and shape their cultural life, I refer to this as “semiotic democracy.” (Semiotic Democracy at 35)

If we can create space where you and I can make things, ambitious things, small things, things that sometimes liberally draw on what has come before, without being immediately told no and without the constant fear that we will be sued into bankruptcy the moment we engage in that act of creation, we will have the ability to shape our creative lives differently. If our laws and norms give us a sphere of protection around our personal uses, in a way roughly equivalent to sketching in a private notebook (even if that notebook is built by a large corporation), a greater percentage of people will express themselves and explore the world more fluidly than they would otherwise.

In Sag’s vision of the future (again, probably a realistic view), memorization leads to guardrails and licensing deals, where we will be told no when a given use seems too risky and a negotiated solution cannot be secured. We will be told yes, probably with caveats, when big moneyed players can forge a path forward. As users, our sense of the possible will be shaped by corporate interests that will often be misaligned with our own.

In the version of the future raised by Geddes, memorization is not something to run away from. The more the models memorize, and the greater the breadth and diversity of the materials on which they are trained, the more users will be able to do with them, and the richer the tools will become. Coupled with robust protections for private, non-commercial use (uses that an individual makes for themselves, their family, or their close friends), AI technologies will not fall into the trap of overcorrecting and blocking legitimate uses, nor be defined and dictated exclusively by commercial interests.

To be clear, the Detective Snoopy case above is easy in both Sag’s and Geddes’s worlds. It’s commercial, public-facing, and substitutional, and it loses. The interesting cases are the ones in between, and that’s where the choice of framework matters: a licensing regime defaults to permission, a personal-use sphere defaults to freedom, where programmers can pursue open source AI projects and authors can use those AI models to experiment with new forms of expression.

What we lose if we accept the framing that says any memorization weighs dispositively against fair use is greater than what we lose to memorization itself. We lose the possibility of a sphere where individuals can read, write, draw, and play with these tools without first negotiating a license. We lose the version of this technology that lets more people participate more fully in cultural life, not fewer. Our works are in the models. We should let them be in the models. Not everybody will get precisely what they want and it can still be fair.

This is the latest in our series of posts marking the 50th anniversary of the Copyright Act of 1976. To find a list of all the posts in this series, click here. This post by guest author, Jonathan Band, is based on a presentation made on March 27, 2026, at the Texas A & M Law School Copyright Act @50 event. The author is grateful to Authors Alliance for its research on the provisions discussed in this post.

Copyright lawyers, whether they are practitioners or academics, treat the Copyright Act of 1976 as if it were a sacred text. To a certain extent, we do so because we have internalized the great deference federal courts generally show to all acts of Congress. The courts give acts of Congress a presumption of legitimacy. The courts assume Congress knew what it was doing when passed a law, and it is not the courts’ role to second guess Congress’s policy judgments or motivations or competence. Moreover, as the Supreme Court teaches, they must “give effect, if possible, to every clause and word of a statute.”

However, I believe that we revere the ‘76 Act even more highly than most statutes because of the process followed in the development of the Act. Starting in the mid-1950s, the Copyright Office produced more than 30 studies on various topics, either internally or by retaining outside experts. The Office then held roundtables, and issued a report in 1961. This was followed by draft legislation in 1963. The action then shifted to Congress, where legislation was introduced and hearings were held in 1964. The process continued in Congress intermittently until enactment in 1976. Thus, the ‘76 Act started with experts and the expert agency, and proceeded in an open and inclusive process. Barbara Ringer, who managed much of this process for the Copyright Office in several different positions, culminating in Register in 1973, is widely considered the architect of the Act, and is viewed as an extremely knowledgeable, dedicated, and fair public servant.

At least, that’s the origin story of our sacred text. I don’t know how inclusive the process really was; I wasn’t there. Nor do I know if Barbara Ringer was as fair and balanced as everyone claims; I never met her. But she is treated in the literature as the Moses, or Muhammed or Joseph Smith of the ’76 Act.

In any event, given the sacred status we accord the ‘76 Act, there is a sense of inevitability around it, not only that it was inevitable that Congress would replace the 1909 Act with the ‘76 Act, but that it was inevitable that each of its provisions would come out the way they did. It is also infallible, and we in the Copyright priesthood parse every word and clause to find its underlying truths.

In fact, there was nothing inevitable or infallible about the provisions of the ‘76 Act. I’ll demonstrate this point by walking through three provisions of the Act that changed dramatically over the two-decade process: termination of transfer; fair use; and statutory damages.

I. Termination of Transfer of Rights

The treatment of the termination of an author’s transfer of rights evolved significantly during the development of the ’76 Act. The original proposal was far more author-centric than the provision Congress ultimately enacted. It would have made it far easier for authors to benefit from the appreciation in the value of their copyrights.

1961 Report of Register of Copyrights. Termination in the Register’s 1961 proposal was inalienable and automatic after 25 years.

1963 Copyright Office Draft Legislation. The Copyright Office’s 1963 draft legislation contained a straightforward provision implementing its 1961 proposal:

In the case of any work other than a work made for hire, a transfer by the author after the effective date of this act, otherwise than by will, of copyright or of any exclusive right under a copyright shall be effective for no more than twenty-five years from the date of its execution. Notwithstanding any agreement to the contrary, at the end of said twenty-five year period the right or rights transferred shall revert to the author or, if he is dead, to his legal representatives, legatees, or heirs at law.

Enactment. By the time it was enacted in 1976 as 17 U.S.C. § 203, the termination language became far more complicated with procedural hurdles. The 1963 proposal was 93 words long; section 203 is 1093 words long. In addition, the ’76 Act contained an equally complex, 1,271 word section addressing pre-1978 transfers, section 304(c). For post-1978 transfers, the earliest the termination could occur was 35 years after the transfer, rather than 25 years in the 1963 draft. These changes neutralized the efficacy of the provision for authors, to the benefit of publishers and music labels.

II. Fair Use

The treatment of fair use also evolved significantly during the development of the ’76 Act. The 1909 Act did not address fair use; it was entirely a judicial creation. Indeed, the Copyright Office considered the alternative of leaving fair use as a common law doctrine, before deciding to codify it.

1958 Latman study. The study Alan Latman conducted for the Copyright Office, considered the fair use case law, various treatments of fair use in the earlier bills which were to replace the 1909 Act, and how foreign countries handled exceptions, particularly the UK’s fair dealing provision.

The study identified four options:

1) maintain statutory silence;

2) recognize the doctrine in broad terms but provide no clarification concerning criteria;

3) recognize the doctrine and specify criteria; or

4) adopt specific exceptions.

Although the study did not express a preference, it concluded that option 3, recognizing the doctrine and specifying criteria, was the “boldest attempt to treat the problem” but the “approach is beset by practical obstacles.” Comments by other experts recommended either silence or general codification.

April 1961 Tentative Report of Register of Copyrights. The draft of the Copyright Register’s report in effect proposed option 2 above, codification of the doctrine in broad terms without specifying criteria.

The statute should include a provision affirming the general principle that fair use does not infringe the copyright owner’s rights.

July 1961 Final Report of Register of Copyrights. The Register’s final report recommended affirming the general principle of fair use as well as an indication of its “scope.”

The statute should include a provision affirming and indicating the scope of the [general] principle that fair use does not infringe the copyright owner’s rights.

1963 Copyright Office Draft Legislation. The Copyright Office draft legislation went beyond proposal in the Register’s final report by including for the first time criteria for evaluating whether a use was a fair use: the four fair use factors. (According to Professors Mike Carroll and Peter Jaszi, it appears that Barbara Ringer herself drafted the four fair use factors.)

FAIR USE IN GENERAL

The statute should include a provision affirming and indicating the scope of the principle that fair use does not infringe the copyright owner’s rights.

§ 6. LIMITATIONS ON EXCLUSIVE RIGHTS: FAIR USE.

All of the exclusive rights, specified in section 5 shall be limited by the privilege of making fair use of a copyrighted work. In determining whether, under the circumstances in any particular case, the use of a copyrighted work constitutes a fair use rather than an infringement of copyright, the following factors, among others, shall be considered:

(a) the purpose and character of the use,

(b) the nature of the copyrighted work,

(c) the amount and substantiality of the material used in relation to the copyrighted work as a whole, and

(d) the effect of the use upon the potential value of the copyrighted work.

1964 bill. The fair use provision in the identical bills introduced in the House and Senate in 1964 differed from the Copyright Office’s draft language primarily by the addition of the legitimate purposes in the “chapeau.”

6. Limitations on exclusive rights: fair use.

Notwithstanding the provisions of section 5, the fair use of a copyrighted work to the extent reasonably necessary or incidental to a legitimate purpose such as criticism, comment, news reporting, teaching, scholarship, or research is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use, the factors to be considered shall include:

(1) the purpose and character of the use;

(2) the nature of the copyrighted work;

(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and

(4) the effect of the use upon the potential market for or value of the copyrighted work.

Enactment. The fair use provision evolved slightly between 1964 and enactment in 1976. The removal of the phrase “to the extent reasonably necessary or incidental to a legitimate” purpose eliminated a potentially ambiguous requirement that the user might have difficulty meeting. The addition of “multiple copies for classroom use” as an example of an acceptable use for teaching was a helpful clarification, while the phrase “whether such use is of a commercial nature or is for nonprofit educational purposes” had a narrowing impact.

Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—

(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;

(2) the nature of the copyrighted work;

(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and

(4) the effect of the use upon the potential market for or value of the copyrighted work.

From the perspective of 2026, it is amazing to think that the Copyright Office actually considered not codifying fair use, and indeed leaned in that direction in 1958. It is impossible to predict what would have become of fair use had it remained entirely a common-law doctrine.

III. Statutory Damages

One of the Copyright Act’s most controversial features is the potentially draconian liability for statutory damages, particularly in cases involving a large number of works. The Copyright Office’s original proposal would have been much more moderate, allowing only one award of statutory damages for all works involved in a single action.

1961 Report of Register of Copyrights. The 1909 Act contained a confusing statutory damages provision with a basic range of $250 to $5,000. The Register’s 1961 proposed authorizing the court to award statutory damages of between $250 and $10,000, as it deemed just, when the award of actual damages and defendant’s profits for all infringements would be between $250 and $10,000. In effect, this would allow the court to ratchet up the damages award when it felt the award for actual damages was insufficient, but the ceiling was $10,000.

1963 Copyright Office Draft Legislation. In response to fierce lobbying from the copyright bar, the Copyright Office proposed a significant increase in the amount of statutory damages available: allowing statutory damages of up to $10,000 per work infringed (i.e., an award of up to $10,000 “for all infringements of a single work for which the infringer is liable”) as opposed to one award for all infringements of all works at issue in a single action. The court would award the copyright owner the larger of actual damages or statutory damages (up to $10,000). The Office proposed limiting the availability of statutory damages to works that were registered prior to infringement.

1964 Bill. This version allowed the court to increase the amount of statutory damages above $10,000 per work infringed in cases where the copyright owner proves that the infringement was committed willfully after service upon the infringer of a written notice to desist. The bill also clarified that the parts of a compilation or derivative work constitute one work.

1965 Bill. This version provided for the reduction of the minimum amount of statutory damages to $100 in cases of innocent infringement. It also placed a $20,000 maximum per work on the award of statutory damages in cases of willful infringement.

1966 Bill. This version added a provision remitting statutory damages to libraries and educational institutions that reasonably believed that their infringing conduct was a fair use.

1969 Bill. This version increased the ceiling on statutory damages to $50,000 in cases of willful infringement.

Enactment. No substantive changes were made to the statutory damages provision between 1969 and enactment. But the journey from 1961 to 1969 was significant. By providing an award of statutory damages on a per work infringed basis, with a substantial multiplier for willful infringement, the ’76 Act set the stage for the copyright wars between the technology and entertainment industries. Congress amended the range of damages twice (in 1988 and 1999) to its present normal range of $750 to $30,000 per work infringed, with a reduction to $200 for innocent infringement and an increase to $150,000 for willful infringement.

Conclusion

There was nothing inevitable about the 1976 Act—that it would pass when it did, and what its provisions would look like. And the text is not infallible. It represents the result of many individual amendments over a two-decade legislative journey. The changes concerning termination benefited publishers and other distributors at the expense of authors; the changes to statutory damages benefited copyright owners generally; and it is hard to predict the impact of a somewhat granular codification of fair use.

The fairness hearing in Bartz v. Anthropic took place this afternoon, Thursday, May 14, 2026, before Judge Araceli Martínez-Olguín in the Northern District of California. The hearing addressed approval of the $1.5 billion class action settlement, Class Counsel’s fee request, and the objections and replies filed by class members and others.

For background on the proposed order, the claims rate going into the hearing, and the principal objections, see our pre-hearing post.

The Court did not rule from the bench. Judge Martínez-Olguín indicated she would take the matter under submission and expects to issue short orders as soon as tomorrow. What follows are our preliminary observations.

Observations from the Hearing

The claims rate has continued to climb. Class Counsel reported that the claims rate, which stood at 91.3% as of the April 30 update, is now 92.77%. Counsel also stated that opt-outs remain modest: 350 valid opt-outs covering 1,802 works, with a handful of late opt-outs received after the deadline. Counsel has asked the Court to require those late filers to show a good reason for missing the deadline before their opt-outs are accepted.

Class Counsel framed the volume of “wrong class” objections as evidence of the settlement’s strength. A recurring theme in counsel’s presentation was that many objectors are not class members at all: they are authors and publishers who want their works included and are unhappy that their works were left off the works list for one reason or another. Counsel argued that this reflects how desirable inclusion in the settlement has become.

Objectors were given only two minutes each. That allocation noticeably constrained the presentations. The objectors who appeared raised, in compressed form, several of the arguments already developed in their written filings:

• Stacy Werner (attorney and author of seven books) argued that the settlement structure gives insufficient attention to independent authors and publishers, and raised the publisher’s failure to register copyright as a distinct issue.

• Victoria Pinder renewed her group-registration objection: four of her works appear on the Works List under a single group registration (TX0008716839), and she will receive a single share for what she describes as separately authored and separately used books.

• Matthew Chase focused on pseudonymous works, arguing that the Works List methodology fails authors who followed Copyright Office practice on pseudonyms and disadvantages self-published authors whose names do not cleanly match the list.

• Natasha Grace raised a similar group-registration issue and asked the Court to consider a remedial process by which authors who believe they have been wrongly excluded could submit their titles for inclusion.

• George Tombs objected to the ongoing use of the works, including the many unregistered works he believes were used in training but fall outside the class.

• Robert Jacobson argued that $3,000 is a small fraction of the $150,000 statutory ceiling and that the absence of information about how individual works were used in training makes proportional analysis impossible.

• Ramsey Hootman asked why a one-time payment is an appropriate resolution if Anthropic continues to profit from the works on an ongoing basis.

James Bartolomei, counsel appearing for objectors, made the argument that the notices about the settlement were insufficient because they didn’t include all relevant information. In particular, he argued that Judge Alsup’s fair use order was never made available to the class through the settlement website, and that the underlying summary judgment briefing was likewise absent. The fair use order is central to understanding the residual liability Anthropic faced — and therefore central to a class member’s decision whether to opt out or accept $3,000. Bartolomei requested a 35-day extension of time on that basis.

Judge Martínez-Olguín’s questioning focused heavily on the fee request and on expense accounting. Most of the bench’s questions concerned attorneys’ fees and the structure of the cost reserve, not the merits of the objections.

• On the service awards ($50,000 to each of the three named plaintiffs), Class Counsel justified them by pointing to the visibility of the case, the work undertaken by the named plaintiffs, and the absence of objections to the awards.

• On the cost reserve (a pool of settlement money set aside up front to cover ongoing costs of administering the settlement), the Judge pressed counsel on why expenses were being handled through a reserve rather than reimbursement, and signaled skepticism about money being administered outside the court’s view. Counsel pointed to Dkt. 619-1 for the cost detail, described a $15 million cap with the lion’s share going to ongoing administration, and represented that unspent funds would return to the class.

• On the actual attorneys’ fees and their fee multiplier, the Judge asked counsel to identify the three best Ninth Circuit cases supporting the size of the requested fee. Some background may help here: in large class actions, courts often cross-check a percentage-of-the-fund fee request (here, 12.5%) against the lawyers’ actual hours worked multiplied by reasonable hourly rates (the “lodestar”). The ratio between the requested fee and that hours-based figure is the “multiplier.” Class Counsel reported a current hours-based figure of approximately $22 million, with $27 million total projected to account for future work; the requested fee is 6.92 times that amount. Counsel argued that the claims rate and the result for the class justify the multiplier, and that Class Counsel’s work in developing the Works List, uncovering the piracy issues, and litigating in an untested area of law warrants the requested fee.

What Happens Next

Judge Martínez-Olguín took the matter under submission. She indicated that short orders may follow as early as tomorrow. We will continue to track this case and post updates as it develops. The full docket is available on CourtListener.

The fairness hearing in Bartz v. Anthropic is set for today, Thursday, May 14, 2026, at 2:00 p.m. Pacific Time before Judge Araceli Martínez-Olguín. The hearing will be open to the public via Zoom. The order recommends logging in early to avoid any capacity or connectivity issues.

Two developments to note before the hearing:

• The claims rate is now 91.3%. In an updated claims report filed April 16, Class Counsel reported claims for 440,490 of the 482,460 works on the Works List (Dkt. 643), up from the 54% reported in the March 19 motion for final approval. Note, however, that we really don’t know what percentage of author or publishers implicated in the suit responded. Because some large publishers hold rights in tens of thousands of works (and presumably those publishers are aware of the suit and made claims) the actual percentage of claimants responding is likely more modest).

• Class Counsel has filed a Proposed Order asking the Court to grant final approval and overrule all objections (Dkt. 646-1; full docket here). The proposed order revises Class Counsel’s fee request downward from 15% to 12.5% of the $1.5 billion settlement fund, plus approximately $3 million in incurred expenses, an $18.22 million cost reserve, and $50,000 service awards to each of the three class representatives. (Proposed order at 12)

See our previous post for background on the types of objections we are likely to hear at the fairness hearing.

What Class Counsel Asks the Court to Do in the Proposed Order

The Proposed Order is a category-by-category roadmap for how Class Counsel would like the court to dispose of objections to the settlement. Whether the court decides to follow this roadmap is an entirely different question, and this hearing will likely help us better understand the likelihood of that outcome. Here are some, but not all, of the elements in the Proposed Order:

• Objections from non-class members are stricken for lack of standing. If the proposed order is followed, this captures most objections arguing the Works List is underinclusive, that particular books should be in the class but aren’t because of the ISBN/ASIN requirement, the U.S. registration requirement, the registration-timing requirement, or because Class Counsel failed to identify them. (See Proposed Order at 8)

• The notice campaign is defended as adequate, citing the 56-day direct notice period reaching to 99.5% of identified rightsholders. Objections that the notice steered class members toward filing claims rather than opting out are disputed (“As to content, the Court-approved Notice expressly stated that statutory damages could range from $200 to $150,000 per Work, and gave “equal dignity” to the options of filing a claim and opting out.”). (Proposed Order, 7-8)

• Late opt-outs are denied with two exceptions. Laura Esquivel and her nephew Jordi Castells, which are denied without prejudice on the ground that they have not shown the consent of other legal or beneficial owners of the six works at issue (“For this reason, Esquivel’s and Castells’s untimely opt-out requests are DENIED without prejudice”) (Proposed Order at 12).

New Replies on the Docket

Several objectors have filed replies challenging Class Counsel’s characterization of their objections. Here’s a sampling:

Victoria Pinder (Dkt. 648) raises an issue with how payouts work across works registered as a group. Pinder is the author of 40+ independently published novels that she registered under a single, lawful group copyright registration. Under the settlement, each copyright registration number counts as one “claimable Work.” An author who registered 40 books individually would receive 40 shares; Pinder receives one — for the same number of distinct books, each separately used by Anthropic, each with its own ISBN or ASIN. She reports that when she raised this with Class Counsel, she was told that counting her individually published books separately could constitute a “windfall.” Class Counsel’s proposed order does not engage with the group-registration question on the merits; it responds that “any author or publisher can register their work on an individual basis,” which Pinder argues is non-responsive.

Chinonto Chakanga (Dkt. 649) argues that Class Counsel’s reliance on Judge Alsup’s notice-remediation orders stands in contrast to what those orders actually say. On Chakanga’s reading, those orders document Judge Alsup’s findings that the original notice was engineered to steer class members away from opting out. Judge Alsup “commanded” equal dignity going forward but did not find it had been achieved. Chakanga also raises a separate due process argument: thousands of pages of sealed summary judgment records, class certification briefs, and expert declarations were not unsealed until January 21, 2026, eight days before the original opt-out deadline and 19 days before the extended deadline. Citing the Ninth Circuit’s recent decision in Avery v. TEKsystems (January 28, 2026), Chakanga argues that compressed timing during the holiday period combined with a sealed judicial record is a substantive defect in the opt-out process.

Laura Esquivel (Dkt. 651), the Mexican author of Como agua para chocolate, challenges Class Counsel’s novel “co-owner consent” requirement for her late opt-out. As her reply notes, the Long Form Notice expressly provides that a single rightsholder’s decision to opt out binds co-authors and publishers; the settlement framework treats opt-out as a unilateral right, not one requiring co-owner consent. Esquivel also argues she meets the Pioneer excusable-neglect standard: she did not receive notice until March 3, 2026 (three weeks after the deadline); she lives in Mexico and reads Spanish, but no Spanish-language notice was provided; and she suffered a stroke during the relevant period.

A Reminder of the Stakes for Authors

For context on what is at stake at the per-author level, the objection filed in January by James Sills (Dkt. 600), reported that Elsevier proposed to take 88% of the per-work settlement amount for one of his books, leaving 12% for the author. The settlement’s default split for non-education works is 50/50, but education works (including textbooks) are subject to negotiation and arbitration governed by the Special Master process.

We do not yet know how representative Sills’s experience is across the author class. But Professor Lea Bishop’s intraclass fairness objection (Dkt. 602) raises the related structural concern that the settlement’s allocation framework, including a Special Master whose career was built representing publishers, forfeiture of unclaimed author funds that will go, in part, to publishers, and the absence of class-counsel assistance to authors in publisher splits, systematically advantages publishers in what is described as a “zero-sum exercise” between authors and publishers for each work. (Objection at 2)

If the Sills 12/88 split turns out to be common among textbook authors negotiating with major academic publishers, the practical recovery for many education-class authors will look very different from the $3,000-per-work figure that has been widely reported in the press.

Attending the Hearing

The fairness hearing is at 2:00 p.m. Pacific Time on Thursday, May 14. Anyone interested in following this litigation is encouraged to attend. Public access details are in the court’s order; again, we recommend trying to log in early.

Post-Hearing Updates

We will provide an update to this post with observations and key takeaways from the fairness hearing as soon as we are able.

Authors Alliance is pleased to announce our collaboration with the eBook Study Group on a joint statement supporting Illinois HB5236, the Digital Library Protection Act, which passed the Illinois House without a single no vote and now heads to the Senate. The bill is a measured response to a major problem: Illinois libraries are spending large amounts of public dollars on ebook license contracts that undermine patron privacy, block ordinary library functions, and make it more difficult for libraries to buy and maintain a ebook collection of works from a broad range of authors.

Our support for this kind of legislation is not new. We wrote in 2021 about the first wave of state efforts in Maryland, New York, and Rhode Island, and in 2022 we endorsed model legislation as a careful response to the copyright preemption questions raised by the Maryland litigation. The current generation of bills, which includes Connecticut’s 2025 landmark law, HB5236 in Illinois, and similar measures in other states, reflects a careful approach that does not dictate prices or compel publishers or authors to license. Rather, they focus on the terms libraries may agree to, ensuring those terms are consistent with libraries’ public mission of lending, access, and preservation.

Allowing libraries to fulfill their public service mission matters greatly to many authors. Libraries are where readers find authors of mid-list and debut books they would not otherwise encounter, and they remain a critical means of discovering authors’ backlist books even after commercial demand has faded.

Public library collection budgets are currently dominated by spending with a small number of publishers who insist that libraries re-license works repeatedly to maintain access and that charge libraries, in some cases, 10x or more the retail price. The result is that many libraries can only maintain access to a small number of high-demand bestsellers, and other authors are left entirely off libraries’ digital shelves. We made this point last month in our letter to the Minnesota Legislature supporting HF 3698 / SF 3685, and it applies with equal force in Illinois.

We will continue to support these efforts as they move through state legislatures, and we encourage authors who agree to make their voices heard. You can read more about these bills and the issues they seek to address on the eBook Study Group’s substack, here.

Today in the Southern District of New York, Elsevier, Cengage, Hachette, Macmillan, and McGraw Hill — joined by novelist Scott Turow and his company S.C.R.I.B.E., Inc. — filed a class action against Meta and Mark Zuckerberg over the use of copyrighted works to train Meta’s Llama models. The complaint has six counts: reproduction by torrenting, reproduction from web-scraped datasets, reproduction during training, distribution by torrenting, contributory infringement against Zuckerberg personally, and removal of copyright management information under DMCA §1202(b).

Much of the underlying conduct alleged in the complaint is already being litigated in Kadrey v. Meta, where individual authors have raised overlapping claims about Meta’s training practices. What makes this new filing distinctive is not really the conduct it targets but the content it is focused on and the class it proposes.

Who and what is swept into the case

The complaint defines the proposed class as all legal or beneficial owners of registered copyrights, in whole or in part, for any book with an ISBN or any journal article with a DOI or ISSN that Meta reproduced through torrenting, scraping, or training. The class is limited to works registered with the Copyright Office either within five years of publication (and before Meta’s use) or within three months of publication.

Importantly, the class covers authors and publishers who own copyright of academic journal articles, in addition to trade books and educational textbooks which are what most previous suits have included. The sample works listed in the complaint’s Exhibit A include five Elsevier journal articles spanning oncology, urology, and the philosophy of mind. One of those articles — a paper titled Monoclonality of Multifocal Epithelioid Hemangioendothelioma of the Liver by Analysis of WWTR1-CAMTA1 Breakpoints, by authors at Memorial Sloan-Kettering in New York — was NIH funded and is freely available to the public on PubMed Central.

On my reading, the class definition would cover academic authors who have published a journal article during the class period and retained their rights. This would include authors who retained partial rights in their work and authors who hold copyright jointly with their publishers.

The class also extends to a much larger body of academic work where authors have formally signed copyright over to publishers — the standard arrangement at most commercial journals, including Elsevier’s. Many of those authors retain specific rights, such as the right to deposit a postprint in a university institutional repository under a Green Open Access policy. But the core exclusive rights at issue in this lawsuit — the right to reproduce copies under § 106(1) and the right to control CMI under DMCA §1202(b) — sit with the publisher, which makes the publisher — not the author — the class member. An author who specifically pushed to preserve public access to their own work would therefore have no standing to opt out of litigation pursued in the publisher’s name, and no direct say in any settlement that results.

Finally, the class definition contains an important wrinkle: it covers only those works that Meta used “without authorization.” Many fully and hybrid open access journals release articles under Creative Commons licenses, most commonly CC BY, which grant broad permissions to copy, distribute, and prepare derivative works on stated conditions, including attribution and, in some variants, limits on commercial use. Whether those licenses authorize what Meta is alleged to have done — copying entire articles into a training corpus to build and commercialize Llama, while stripping copyright management information in the process — is a question that has not been squarely resolved.

If sorting class members from non-class-members requires the court to decide which CC-licensed works fall inside the class definition and which fall outside, this case could end up being a vehicle for resolving whether and how common open licenses cover AI training. That is a question with implications well beyond Meta and Llama. It also places OA publishers, and the many authors who chose CC licensing precisely because they wanted their work used and reused, in the awkward position of having their license terms construed in a class action they may not even know they are part of.

What happens next

Class actions require class representatives to fairly and adequately represent the class. I think many academic authors in this class will not be enthusiastic about Elsevier representing their interests in court (I’m definitely one of them), and the class representatives may have a hard time showing that they can do so in light of a history of an antagonistic relationship with many members of this class. This would include, to give just one example, the large number of academic authors who have publicly boycotted Elsevier in response to its business practices.

But the case isn’t so straightforward — I think many are also wary about Meta as a company, particularly as this case may lead toward a settlement that would allow Meta to train LlaMa on licensed Elsevier content. Academic authors and researchers stand to gain little from such settlement or licensing deals, and may stand to lose much if the fair use defense is eroded for web scraping or computational analysis in the course of this litigation.

For now, this is a putative class action, and a lot has to happen before it goes anywhere. Meta will respond, motions will be filed, and at some point the court will decide whether to certify the class. If certification happens, notice will go out to class members, and individual rightsholders may have an opportunity to opt out.

Authors Alliance will be following the case closely and will share more as it develops. For now, the most useful thing scholarly authors can do is recognize that many of their works will be implicated in this suit and take steps to stay informed as it progresses.

Authors Alliance submitted a comment to the US Copyright Office in response to its proposed fee schedule (Docket No. 2026–2). We focus our comment on two things: (1) commending the Office for imposing heavier fee increases on corporate users—such as those registering large photographic and non-photographic databases, and (2) urging the Office to stay true to its stated policy goals of incentivizing creativity by not hiking up the price of recording termination notice.

Copyright termination of transfer is a statutory right under Section 203 and Section 304 of the 1976 Copyright Act. It allows authors to reclaim ownership of works they previously signed away. When a novelist or academic writer transfers their copyright to a publisher, that transfer isn’t permanent. After a waiting period of 35 years, the author can serve a termination notice and take back the rights, regardless of what the original contract says. The right is inalienable, meaning no contract clause can waive it in advance. Congress built it that way deliberately, recognizing that creators often negotiate from a weaker position compared to the distributor. Termination gives authors a chance to license on better terms, or simply reclaim control over work that is central to their creative legacy.

Recording the termination notice with the Copyright Office is the formal step that is required to make termination legally effective (See our step-by-step guide on how to file a termination notice). That is why the fee the Office charges authors to do so matters enormously to the human authors.

We flag three key problems with the proposed fee for recording termination notice:

First,we point out the fee for recording termination is proposed to increase a staggering 126%, compared to the “average of 43%” increase in all registration and recordation services. We urge the Office to cap the fee increase for recording termination notice to the “average of 43%” increase in its services, making the fee for recording termination notice no more than $135.85 for electronic and $178.75 for paper. We also urge the Office to consider the possibility of further limiting the fee increase to account for the fact that many authors are already priced out of exercising termination rights.

Second, we point out that recording termination notices require 8.5x the expenditure in staff training and similar expenses as registering a new copyright. For comparison, registration entails labor-intensive review by trained professionals, yet standard registration for a literary work with e-deposits—the most costly category of work out of all standard registrations—has a direct cost of $21.37 and indirect cost of $125.14 per registration, or $146.52 in total. Yet recording termination notice is reported to incur a direct cost of $95.68 and indirect cost of $1,006.88, totaling $1,102.56 for each electronic submission. We ask the Office for clarification.

Third, we point out it is not clarified if counter-notice processing costs are being folded into the fee for recording termination notice, and whether authors alone are made to bear those costs. We do not think the Office should ask authors to bear the cost of processing counter-notices. Termination was introduced by Congress as a more powerful safeguard for authors than the renewal system that existed before; therefore it should never be made financially more burdensome to exercise. Furthermore, intermediaries are incentivized to oppose termination notice even without the counter-notice fee exemption.

The Office is also soliciting comments on a new fee structure. We encourage you to share your ideas with the Office.

In May 2024, we filed an amicus brief urging the Tenth Circuit to rehear Whyte Monkee Productions v. Netflix. The previous decision was disastrous for documentary filmmakers, historians, and anyone else who uses photographs, movie clips, or other works as historical reference points or for illustration. That’s because the court previously took the position that fair use generally disfavored uses that didn’t directly comment or criticize creative aspects of the underlying work. This meant that all sorts of common uses that engaged with the subject of an underlying work rather than its artistic content were now in jeopardy. It came to this conclusion based on what was, in our view, a significant misreading of the Supreme Court’s recent guidance in Andy Warhol Foundation v. Goldsmith.

Thankfully, the court agreed to rehear the case. Yesterday, it issued a new opinion, and it completely reversed course. The decision affirmed the lower court’s holding that Netflix’s use—where it took a short video clip from a recording of a memorial service to give context about the main character in its documentary series—was fair use.

As Yuanxiao Xu discusses here, there are only a few post-Warhol fair use appellate cases, and Whyte Monkee was one of the more concerning because it took a particularly constrained view of what “transformative use” means, potentially limiting its use for creators in the future. The new opinion largely corrects that.

What the court says about transformative use after Warhol

Unlike the actual show, the facts of the case are straightforward: Timothy Sepi shot a video of the funeral of Joe Exotic’s husband, Travis Maldonado, which was livestreamed on YouTube. Netflix used, without permission, about 66 seconds of that footage in Tiger King, juxtaposing Joe Exotic’s eulogy with critical commentary from Maldonado’s mother. The district court found fair use; the original Tenth Circuit panel reversed, concluding that the use was not fair.

The key issue in both the first opinion and this new one is whether Netflix’s use is viewed favorably under the first fair use factor, which evaluates the purpose and character of a use. As part of that analysis, courts look at whether a use was “transformative,” which means asking “whether the new work merely ‘supersede[s] the objects’ of the original creation . . . (‘supplanting’ the original), or instead adds something new, with a further purpose or different character.” This has been the basic formulation of “transformative use” since the Supreme Court adopted it over 30 years ago in Campbell v. Acuff-Rose.

In Warhol1, the Supreme Court revisited transformative use. As that court explained, “Whether a use shares the purpose or character of an original work, or instead has a further purpose or different character, is a matter of degree. Most copying has some further purpose, in the sense that copying is socially useful ex post. Many secondary works add something new. That alone does not render such uses fair. Rather, the first factor (which is just one factor in a larger analysis) asks “whether and to what extent“ the use at issue has a purpose or character different from the original. The larger the difference, the more likely the first factor weighs in favor of fair use. The smaller the difference, the less likely.”

So, how does one evaluate how much is enough? In the first version of this case, the 10^th circuit didn’t get beyond the facts of Warhol, essentially concluding that unless a use targeted underlying creative or artistic aspects of the original—criticizing or commenting on it, which is what the court in Warhol looked for in that case—then whatever it was, it was simply not transformative enough.

This time around, the 10^th Circuit took a closer read of Warhol and focused in more precisely on the emphasis Warhol places on “justification” for using the original work. In doing so, it identified two different pathways to showing “justification.”2

For the first pathway, the court (quoting Warhol) explains a secondary work that “has a distinct purpose is justified because it furthers the goal of copyright”—viz., “to promote the progress of science and the arts, without diminishing the incentive to create.” “If the secondary use is transformative in this broad sense, then no independent justification is likely to be necessary.”

The second pathway is narrower — a secondary work that shares the same or similar purpose as the original work is more likely to be a “substantial substitute” for the copyrighted work — “which undermines the goal of copyright.” Accordingly, it is unlikely to be deemed justified in the broad sense. If such a use is to be justified, it likely must be in a “narrower sense.” In this narrower sense, “a use may be justified because copying is reasonably necessary to achieve the user’s new purpose.” (slip op. at 28, citations omitted).

The court holds up parody as a paradigmatic narrower-sense case explaining: “The Warhol Court elaborated by pointing to parody as an example where the secondary use of the copyrighted work may be justified in this narrower sense because of the essence of that use: that is, the essence of parody is to target the original work for imitation to achieve some comic effect — specifically, in order for the user to effectuate its parody purpose, the user needs to target the original work.”

Targeting is not required

One major takeaway here is that targeting creative aspects of the original isn’t always required. The most direct rebuke to the narrow reading of Warhol comes when the court addresses the plaintiffs’ argument that Tiger King could not be transformative because it did not comment on or “target” the funeral video itself. The court squarely rejects this: “Although the Supreme Court recognized in Warhol that targeting can be important, the Court crucially stated that ‘targeting is not always required.’”

In addition to this clear statement quoting from Warhol itself, the court also relies on Google v. Oracle, where the Supreme Court found a transformative use of copyrightable software code without any targeting at all — because the secondary use served a different purpose (interoperability and the application of preexisting programmer skills in a new environment). It also cites the Second Circuit’s recent decision in Romanova v. Amilus and the Eighth Circuit’s decision in Griner v. King for the same proposition.

For this particular use, the court makes clear that because the use has such a distinct purpose — Defendants used the funeral video, meant for memorial purposes “to further a larger narrative that, among other things, commented on and educated viewers about Mr. Exotic and the “big cat world” — additional justification isn’t necessary. It is justified “in the broad sense.” But even in other “narrow sense” cases, the court is clear that targeting is just one way to satisfy the need for justification — useful when the secondary use shares a purpose with the original. It is not a generic threshold that all transformative uses must clear. The plaintiffs’ contrary reading, the court says, “reflects a misreading of governing precedent, and we reject it.”

For anyone working in fields where the secondary use is meaningfully different from the purpose of the original but does not involve commentary on the original — text and data mining, AI training on copyrighted works, indexing, search, accessibility, scholarly reuse, and documentary filmmaking — this matters a great deal.

Documentary use is strongly, but not categorically, favored

The court’s treatment of documentary use is notable. The opinion devotes a full section to documentary practice, drawing on two important documentary film fair use cases: Bouchat v. Baltimore Ravens and Elvis Presley Enterprises v. Passport Video.

The court explains that documentaries frequently incorporate snippets of copyrighted material to advance uses that fair use favors, as listed in the statutory preamble — criticism, comment, scholarship, news reporting — and that courts have, accordingly, often found the first factor to favor fair use in documentary cases.

“This outcome does not stem from the fact that documentaries are categorically favored. They are not. There is no presumption in favor of fair use that flows from attaching the label ‘documentary’ to a secondary work.” Rather, the reason documentary uses tend to win on the first factor is because they tend to exhibit the characteristics that matter: an objectively different purpose from the original, incidental or limited use of the borrowed material, and alignment with the purposes listed in fair use’s preamble.

The court uses Bouchat as one example. There, the Fourth Circuit considered the NFL’s use of the original Baltimore Ravens logo — itself the subject of a long-running infringement suit — in a series of commercial videos recounting the team’s 1996 draft, key moments in franchise history, and individual player careers. The court found the use transformative because the logo, originally a brand symbol, was now functioning as a piece of factual historical record — used “for a fraction of a second” in some shots and “for a longer stretch” in others, but always to tell a story about the team rather than to identify it. That the logo itself was unaltered did not matter; the use was animated by a different purpose and was, in the Fourth Circuit’s words, “exceptionally insubstantial.” The label “commercial” did not sink it.

Elvis serves as the cautionary example. There, the Ninth Circuit found the first factor to weigh against fair use because the documentary used substantial portions of the clips for the same intrinsic entertainment value the original works provided. The label “documentary” did not save it.

Applied here, Tiger King‘s use of 66 seconds out of a lengthy funeral video — interspersed with critical commentary from Maldonado’s mother, deployed to illustrate Joe Exotic’s showmanship — has a clearly different purpose from a video shot to memorialize a friend.

As one person pointed out to me, the court’s explanation may be a bit incomplete — one additional area the court didn’t touch on much, but a good reason documentary use may be favored is that they are often relying on prior works for their unprotected factual content versus creative expression. But on the whole, I think the explanation is helpful, and largely aligns with the fair use best practices that documentary filmmakers actually use day to day — limited, purposive use of archival material to tell a different story — without writing a blank check for any work that calls itself a documentary.

A few additional notes

A few smaller features of the opinion are worth mentioning.

First, the court cites the new Restatement of Copyright. To my knowledge, this is the first appellate court citation to it. The court relies on it for a number of important conclusions about how to judge the likelihood of harm to the market for the original. Whatever one thinks of the Copyright Restatement project (I acknowledge it has had some noisy detractors, but I think it has been a worthwhile project with a valuable output), it is now influencing copyright law in real cases.

Second, the court makes some interesting statements about the burden of proof for defendants who may be in a position to challenge market harm. Typically, once the plaintiff demonstrates the existence of some market, the burden is on the defendant to disprove harm (difficult to do, and costly if you need to bring in expert witnesses). Here, the court offers this: “We think that in some cases, akin to this one, the implausibility of substitution can be so apparent that evidence concerning market impact that is extrinsic to the uses of the original and secondary works themselves (including sophisticated empirical evidence, offered by an expert or otherwise) is unnecessary.” Given the challenge of proving a negative, this is a welcome precedent for fair users.

Finally, the court’s reliance on legal scholarship about the meaning of Warhol is notable. It repeatedly cites an article by Shyamkrishna Balganesh & Peter S. Menell for takeaways from Warhol on transformative use and to another by Karen Shatzkin & Dale Cohen on application to documentary film. The reliance on Balganesh & Menell was surprising to me — the takeaway I had from that article would not have led me to the conclusion that the court came to, about how uses with a sufficiently distinct purpose do not need additional justification. But perhaps I need to re-read it more closely!

If you search, you will find that Warhol has resulted in innumerable law journal articles. It’s really overwhelming. You can find articles on how Warhol applies to AI, art in general, appropriation art, art reference uses, documentary films, the introduction of competition analysis in fair use cases (actually one of my favorites and worth reading), and the list goes on.

I know I’m excluding many good ones here, but I’ll end by giving just two recommendations on further reading. First, given the clear importance of “justification” under the first fair use factor, I recommend Pamela Samuelson’s Justifications for Fair Use which the Second Circuit in its post-Warol Romanova v. Amilus decision relied on to explain the many different kinds of cases, outside of criticism and commentary, where courts have found uses to be justified. The second is Michael Carroll and Peter Jaszi’s The Triumph of Three Big Ideas in Fair Use Jurisprudence, which offers, in my opinion, one of the best syntheses of how to read Google v. Oracle and Warhol v. Goldsmith together in light of the broader body of fair use case law.

Share

We are releasing today the final report of the Public Interest Corpus project. A stable, citable version is here: The Public Interest Corpus: A Framework for Implementation. Because we want to encourage additional feedback, we’re also sharing a version here that allows you to comment.

The report is the product of more than a year of work supported by the Mellon Foundation, in which we asked how research libraries can make books data available for AI training and computational research in ways that serve the public interest, rather than reinforcing the existing concentration of access to texts among a small number of well-resourced commercial actors.

We’ve talked about the starting point of this project before: books are now widely recognized as among the highest-quality training data available for AI systems—they reflect sustained editorial processes and capture historically deep records of human thought across many disciplines and languages. Research libraries hold them at scale, often already in digital form. But access to that data for AI development has so far been shaped largely by which commercial actors have the resources to license, scrape, or otherwise acquire it. Academic researchers and smaller public interest organizations have been mostly outside that conversation.

The project was designed to ask whether libraries could change that, organized around their public interest mission and operating within the legal frameworks that already govern their work. Over the course of 2025, we engaged with researchers, librarians, authors, publishers, and technologists through workshops at Northeastern, NYU, and the University of California Office of the President, and through presentations at CNI, the Charleston Conference, AI4Libraries, and other venues. The report distills what we learned about services, governance, legal strategy, and sustainability. A few key takeaways:

• Stakeholders consistently asked for a service oriented toward access for academic and nonprofit researchers, with programmatic bulk access for computational work and discovery-oriented access for students and members of the public.

• Rather than build something new from scratch, the report recommends partnering with an existing academy-owned research service—HathiTrust, for example, is a natural candidate given its mission alignment, infrastructure, and established collaborative relationships across the research library community.

• Coordination of sourcing across institutions, e.g., on engagement with commercial digitization partners, should sit within that framework, drawing on lessons from earlier work such as under the Google Books project, where terms have sometimes constrained downstream library service development more than necessary.

Throughout the project we gave special attention to legal issues, particularly in light of the large number of AI copyright suits pending. From cases like AV ex rel. Vanderhye v. iParadigms (2009) (plagiarism detection tool) through Authors Guild v. HathiTrust (2014) and Authors Guild v. Google (2015) courts have recognized non-expressive computational uses of texts as transformative fair use, with text and data mining specifically identified as the kind of research application fair use is meant to enable. The 2025 decisions in Bartz v. Anthropic and Kadrey v. Meta extend that foundation into the AI training context. Bartz in particular reinforced and extended what the Google Books court concluded about the legality of converting lawfully held print copies into digital form. Those cases paint an encouraging picture for fair use as applied to AI training, but Thomson Reuters v. ROSS Intelligence (2025) is an important counterpoint, concluding that copying for AI training was not fair use in a case focused on direct commercial substitution of competing legal research tools. But even there, the focus was tightly constrained to competing commercial products. Our recommendation is that the Public Interest Corpus, focused on non-commercial research and scholarly uses, operate squarely within this established line of cases, while exercising additional care where specific features of a use—e.g., contractual constraints from prior digitization agreements, downstream commercial applications, verification of independent researchers—warrant heightened attention.

This report closes the planning phase. What comes next is implementation, which would include identifying the right institutional home and sustaining engagement with authors, publishers, and researchers about how this resource should evolve. None of this is novel work for libraries. What is novel is the scale and the moment, and the opportunity to ensure that the next generation of AI tools is built on a foundation that reflects a focus on supporting research, scholarship and learning. To do all of this requires resources, and we’re currently working to secure funding to get this started (stay tuned for more in the coming months).

We are grateful to the Mellon Foundation for supporting this work, and to the many advisory board members, workshop participants, and interview subjects who shaped the report along the way. Personally, I am grateful to my co-PI Dan Cohen at Northeastern, to Thomas Padilla (now at the University of Nebraska-Lincoln) and Giulia Taurino for their great work on this project. You can visit the project website at publicinterestcorpus.org.

This is the latest in our series of posts marking the 50th anniversary of the Copyright Act of 1976. To find a list of all the posts in this series, click here.

When we launched this series, I hinted that, though we would be reading through Title 17 mostly in order, we would occasionally break the sequence and sometimes would combine multiple sections where it made sense. In this post, I'm doing both, skipping ahead to Chapter 10 of Title 17, where we find the Audio Home Recording Act of 1992 (AHRA). Among Title 17's fifteen chapters, Chapter 10 may be the most . . . quaint. It isn’t even really a copyright law. It is almost certainly the last place one would expect to turn at this point in the series. But I have a reason, which I'll get to later in this post.

📀 Track 1: A Day In The Life, by Generation X.

But first, some background. If, like me, you are a child of the 1980s, then you will remember the mixtape. The closest modern equivalent is a shared streaming playlist, but mixtapes were a great deal more. Listening to music on the radio, or on commercial stereos, we compiled cassette tapes of favourite songs, classic advertisements, radio call signs, and air checks from our favourite announcers, and then added often meticulous track information and even elaborate label art to create one-of-a-kind multi-media collages. For many mid- to late-Gen Xers, it was our defining communications tool. We shared mixtapes with our friends (even our enemies). We shared them in moments of joy and of sadness. Mixtapes were often how we flirted, broke up, and apologised, and especially how we tried to cope. If we didn't know how to say it (and at that age, few of us did), we could turn to a mixtape.

Mixtapes raise several copyright questions. By definition, they contain multiple unauthorized reproductions and distributions, and arguably constitute unauthorized derivative works. Whether they are copyright infringements was a point of contention, but it was not an urgent question because the mixtape was made possible by the cassette recorder, which was an analog recording device. Since each reproduction was from an analog source to an analog destination, each successive reproduction was inferior in quality. Compact disc (CD) players, introduced in 1980, allowed for a digital source, but the destination was still analog, since few consumers had the professional model equipment to record to CD. Mixtapes had obsolecense—known in preservation circles as “generation loss”—built-in to their structure. So the recording industry looked the other way.

Everything changed in 1987, when Sony introduced the Digital Audio Tape, or DAT. The advent of DAT recorders signalled the possibility of widespread digital-to-digital reproduction by average consumers. No analog element meant no generational loss, which meant the physical limitations of analog personal recording devices vanished. The possibility now existed for consumers to spread thousands of perfect copies through the market, obviating the need for purchasing commercial copies. With DAT, the perceived threat of consumer recording went from negligible to existential.

Stakeholders looked to Congress for a fix, but it was not a simple matter. In the face of increased infringement (either actual or feared), the first proposal is often additional enforcement, but here that was unworkable. I don’t think it is an exaggeration to estimate tens or possibly hundreds of millions of people were in the habit of using cassette recorders to supply curated playlists for themselves in their daily lives. Many of those uses were undoubtedly fair, but the extent of fair use in that context was not settled. Congress did not relish the prospect of imposing laws that replaced the established norms of millions of Americans with infringement actions and subjected them to many thousands of dollars in statutory damages. But no one wanted the new technology to threaten the recording industry, either.

Given the predicament, Congress tried something new—its first foray into using collective funding, rather than private enforcement, as the balancing force.

📀 Track 2: Chapter Ten, by The Usual Suspects

The Audio Home Recording Act created a levy on consumer digital recording equipment, which raised funds to be distributed to creators and other interested parties. In exchange, consumers received permission to use those devices for personal purposes. To control the downstream spread of digital copies, Congress required all covered devices to include the Serial Copy Mangement System (SCMS), a technical limitation on digital copies designed to mimic generational loss. Compliant DAT recorders could make first generation lossless copies of original digital recordings, but had embedded locks preventing them making second generation copies.

Since I’m collapsing ten sections into a single blog post, here is a quick tour of the AHRA. The Act creates Chapter 10 of Title 17, and divides it into four subchapters.

Subchapter A (Sec. 1001): Provides definitions for terms used throughout the chapter. The definitions are pretty clunky, but the most important term turned out to be the definition for “Digital Auido Recording Device,” which was limited by definition to devices whose primary function was creating music recordings for personal use. The AHRA was unconcerned with spoken word recordings or with other forms of digital data.

Subchapter B (Sec. 1002): Imposes the SCMS requirement on device manufacturers, and includes provisions for verification and enforcement of the requirement. Sec. 1002 also prohibits circumvention of the SCMS system, making the AHRA Congress’s first technical protection measure, a concept which reappeared in a broader form in the Digital Millennium Copyright Act of 1998.

Subchapter C (Secs. 1003–1007): Establishes the royalty fund, and requires device manufacturers or importers to pay 2% of the sale price of the device (a total of at least $1 and no more than $8 per device) into the fund. It also provides the formula for distribution of the funds to “interested copyright parties” (defined in Sec. 1001), which are: the composers and publishers of musical works; the owners of sound recordings; the featured artists on those recordings; and the industry groups that represent them.

Subchapter D (Secs. 1008–1010): Finally, Subchapter D provides for enforcement, and defines potential statutory damages under the AHRA. It also provides the users of AHRA-compliant “Digital Music Recording Devices” with immunity from lawsuits under the Copyright Act for noncommercial uses, and releases manufacturers and distributors from secondary liability.

Chapter 10 is the first sui generis law we’ve covered in this series. If something is sui generis, it is separate and self-contained, in more or less the way that an ad hoc committee differs from a standing committee. The AHRA interacts with copyright law, but it is outside the system established by the 1976 Copyright Act. This is why the AHRA is not, strictly speaking, a copyright law. It is related to copyright, and so is included in Title 17, but it is in its own class and interacts with the copyright provisions of Title 17 (such as duration, exceptions, etc.) only where it specifically says so, rather than by default.

📀 Track 3: Computers Killed the AHRA Star, by The Market Force

If you were alive in the 80s and are wondering how you missed the saga of DATs, it may be because DAT as a format was short-lived and never became popular in consumer-level products (speaking as a former special collections librarian, it was also a terrible format). When recordable CDs (CD-R and CD-RW) were introduced, they quickly supplanted DAT.

The AHRA applied to CDs and audio CD recorders, too, but unlike DAT, recordable CD’s usefulness extended far beyond audio recording. They were highly useful, for example, for storing computer data. But digital audio is simply digital data in an audio format, and so users could also use computers to record digital audio in digital formats (DAT could store data, too, but was never widely used for that purpose). As a result, consumers had two choices for digital-to-digital audio recording: they could use Digital Audio Recording Devices (as defined in Sec. 1001), such as dual CD stereo components, or they could use their computers. If they wanted to use their computers, they could use any CD-R; if they wanted to use dedicated recorders, they needed to buy “Music CD-Rs,” which were the same thing but with SCMS encoded in them.

Though physically no different, Music CD-Rs were much more expensive because their price typically included the cost of the royalty fund levy. My recollection is a spindle of five Music CD-Rs cost roughly the same as a spindle of 50 data CD-Rs. The difference did not stop there, though, because stand-alone CD recorders, which could only use Music CD-Rs, were limited to lossless, uncompressed recording formats, which allowed roughly fifteen perfect tracks per disc. By contrast, a computer could record using any available format, and using lower sampling rates, allowing roughly 200 lower-quality but passable tracks to fit on the same disc. Meanwhile, SCMS-enabled recording equipment was usable only for audio, but computers, of course, had multiple uses.

Consumers generally prefer lower costs to higher ones, and simpler paths to more complicated ones, and they will often choose simpler, less-expensive solutions at the expense of quality. That is overwhelmingly how they ended up choosing their digital audio. The music-as-data path effectively destroyed the need for SCMS-enabled products. The Recording Industry tried to recover ground by forcing computer and data CD-R manufacturers to pay the AHRA levy, but the Ninth Circuit, in RIAA v. Diamond, halted that effort when they ruled that MP3 players—tiny computers designed for playing digital music files—were not “Digital Audio Recording Devices” as defined by the AHRA. Eventually, the system collapsed. The royalty fund, which in 2000 had over $5 million, had been completely disbursed by 2015, effectively ending the AHRA experiment.

📀 Track 4: Back to the Future, by The Past Lives

We are once again in an era where we must decide how to adapt to a profoundly disruptive technology. Like DAT in 1987, the release of generative large language models in 2022 has raised fears that the disruption will cause the collapse of long established business models, and especially in the creative industries. Those fears—many of them understandable—have given rise to a variety of proposed solutions. In addition to the roughly 100 ongoing lawsuits, one of the proposals I’ve seen floated lately is for a levy, to be imposed on AI companies and distributed to creators.

At the beginning of this post, I promised a reason for picking up the AHRA at this point in our series, and here it is: we tried a levy system once before, and it failed, spectacularly. That does not mean a levy couldn’t work in the future—it very well might. But a successful levy would need to learn from the mistakes of its predecessor. Here are three lessons I think a future Congress could learn from its past failure.

Laws work better when they are fully informed by technology. I continue to marvel at how well the Copyright Act of 1976 anticipated future technology, and it did so because Congress engaged in expert-led study that was informed by experts. The AHRA provides a useful contrast—a cautionary tale for laws that are ignorant of the technology they seek to regulate. Once sound is converted to bits of digital data, it can be interpreted and manipulated by anything that can process bits of digital data, whether it is a professional sound component, a laptop computer, or a refrigerator’s internal temperature regulator. And with the right equipment, any such device can share that data with any other such device. “Digital” is a property of the data, not the device.

All of this was well known in 1992. The AHRA regulated specific devices on the assumption the music data would never leave the music ecosystem, which any technologist could have told them was unrealistic. Congress didn’t do its homework, and so the AHRA was doomed to fail. Policy makers looking to legislate around AI would do well to focus on understanding the technology, rather than on understanding the politics of the technology.

Laws driven by intermediaries are unlikely to benefit authors. As soon as a large pot of money appears, there are inevitably parties that are eager for a share, and who will spend heavily to get it. Very often they are parties that are entitled to some of the money, but then use that claim as a wedge to claim an expanded share.

The purpose of the AHRA was ostensibly to benefit creators, but the big winners were the intermediaries—the record labels and the music publishers. Under Sec. 1006, roughly 38% of the royalty fund was distributed to the recording labels—all six of them—and another 17% was distributed to music publishers. The rest was distributed among probably thousands of composers and performers, most of whose share probably was so small as to not be worth claiming. This is to say nothing of the complexity of ensuring equitable distribution among those thousands of creators, by collectives that may not have the creators’ interests at heart.

With pots of more than a billion dollars now at play to compensate creators for the work of AI companies, the courts and Congress should be careful to prioritize the creators themselves and not the louder, better-organized intermediaries.

Laws should be understood before they are changed. In the industry’s digital music panic, mixtapers were never a threat, or a problem. Nor were libraries that lent CDs (many of which certainly ended up being copied), nor friends loaning CDs to other friends, nor teachers making copies for their students to study. As often as not, creators were the beneficiaries of such sharing. The AHRA provided an immunity from suit for some uses, but many of those uses were also never infringements in the first place.

The legislative history shows that Congress spent decades in measured, careful deliberation and consultation with experts before they finally completed the 1976 Copyright Act. One of the things they took pains to establish in that Act was the limits of copyright. As I discussed in an earlier post, exclusive rights are only half of the copyright picture; and the other half, constituting the public domain and fair uses, must be given equal weight. As we have often written, authors need both halves in order to keep creating.

A levy in exchange for an exemption was a novel idea, and it had some merit in theory. In practice, I think the main lesson of the AHRA is that a future levy scheme is probably ill-advised. But regardless, before Congress considers changing the law, it should make sure it understands the law it is changing. Congress should scrutinize who will actually benefit from, and bear the costs of the change, and should take particular care not to inadvertently limit the commons on which authors depend for inspiration.

Correction: In this post, we noted that objections had been unsealed, pursuant to Judge Martínez-Olguín's order ("The Clerk is directed to unseal the following correspondence from objectors: Dkt. Nos. 425, 539-546, 549-552, 564-569, 584-585, 588-589, 593-612.") As of April 20, 2026, several of the objections listed in the order remain sealed. At minimum, Dkt. Nos. 544, 596, 598, 600, 601, and 602 appear to still be sealed. The reason for this is unclear.

On April 8, Judge Martínez-Olguín issued an order to move the Bartz v. Anthropic settlement fairness hearing to May 14, 2026 at 2:00 p.m.

Class members who have timely objected and wish to be heard may join the hearing via the Zoom log-in information found on page 2 of the order.

Participants are advised to log-in early to avoid any capacity or connectivity issues.

For anyone who might want to attend in person, the hearing will be on May 14, 2026, at 2:00 PT at the San Francisco Federal Courthouse, 450 Golden Gate Ave., Courtroom 12 – 19th Floor.

In response to a motion from objector Professor Lea Victoria Bishop, Judge Martínez-Olguín also ordered that Professor Bishop’s objection and those of other objectors be unsealed. This has been done and the objections can be found in the docket. We are glad to see this increased transparency around objections to the settlement; class members deserve to know what other authors found troubling about this settlement before the court decides whether to approve it.

Below is a non-exhaustive list of what appear to be the most common objections found in those documents (I’ve provided a link to an objection for each as an example and reference; given that there will often be more than one objection filed raising many of these issues, these links to individual objection filings are not meant to capture all associated objections):

1. Exclusion of foreign/non-US-registered works from the class [Doc 630 (Dkt. 630), pp. 5-7].

In her objection, Professor Bishop notes that foreign/non-US works need not register copyright in the United States in order to file suit and that this group could represent over 2 million works, a significant large-scale exclusion from the class.

2. Distribution plan systematically favors publishers over authors (the claims-made process that burdens authors and benefits publishers, the arbitration mechanism that disadvantages authors post-settlement, and the presumption that publishers share in authors’ recovery) [Doc 630 (Dkt. 630), pp. 2-6]

For a lawsuit that was brought by authors ostensibly for authors, it’s true that publishers are likely to benefit substantially from this settlement. Professor Bishop notes that ultimately publishers could claim roughly 50% or more of the total settlement amount, which is a profound departure from an author-centric lawsuit.

“Even if 60% of works have been claimed, it may be the case that 55% were claimed by publishers and only 5% by authors. The court may wish to consider scrutinizing data on the percentage of authors that have submitted claims. It is particularly concerning that the distribution plan fails to dedicate unclaimed amounts to escheating or charity. Instead, amounts not claimed by authors can be collected by publishers.” (Objection at 5)

3. Misleading/coercive class notice, specifically the “get nothing” framing that obscures the opt-out value (potential statutory damages of up to $150,000/work vs. ~$3,000/work in the settlement) [Doc 571 (Dkt. 571), pp. 1-4]

The objection raised here is that the class notice did a poor job of making class members aware of their other options, instead leading some class members to believe that the choice was between accepting the settlement or “getting nothing.”

4. Inadequate compensation relative to the actual scope of infringement and statutory damages available [Doc 571 (Dkt. 571), p. 2]

Statutory damages per infringement can technically be up to $150,000/work, so this objection is based on a view that the settlement amount per work is woefully inadequate compared to that possibility.

5. Class notice not received / received too late to meet the opt-out deadline [Doc 635 (Dkt. 635)]

“The “Notice of $1.5 Billion Proposed Class Action Settlement Between Authors & Publishers and Anthropic PBC” was not received by Objector Esquivel until approximately March 3, 2026, almost a full month after the deadline to opt out of the class settlement had expired.”

6. Inadequate representation / Class Counsel conflicts of interest (Judge Alsup’s December 2025 concerns about fee-sharing concealed from incoming judge) [Doc 630 (Dkt. 630), pp. 2-4]

Here, Professor Bishop takes issue with an apparent exclusion of some of Judge Alsup’s concerns from updates to Judge Martínez-Olguín (“Class Counsel’s omission of this record from its January 2026 Status Update and March 2026 Motion to Approve is not an oversight. It is an affirmative misrepresentation of the record to mislead the incoming judge.”)

7. Group copyright registration undercounting (multiple distinct books collapsed into one “claimable work” when registered under a single group registration) [Doc 641 (Dkt. 641), pp. 1-5]

“Yet under the settlement’s methodology, all 40+ of these books are collapsed into a single “claimable Work” simply because I used a single group copyright registration number. The result is that I am compensated as if Anthropic infringed one book, when in fact Anthropic downloaded and used the content of over forty of my books to train its AI model.”

8. Settlement sets a dangerous precedent by allowing an AI company to escape liability for mass infringement at a steep discount [Doc 640 (Dkt. 640), p. 1]

“This settlement allows a multi-billion dollar AI company to “buy” its way out of massive piracy for a “discounted” rate, setting a dangerous precedent that it is cheaper to steal human IP and settle later than to license it fairly.”