Friday’s Endnotes 07/11/25

By , July 11, 2025.

Top Noteworthy Copyright Stories from June 2025 — June was a busy month for copyright. The Copyright Alliance’s Rachel Kim reviews some of the key developments that took place.

UMG Chief Digital Officer on AI Music: ‘If You Don’t Claim a Seat at the Dinner Table, You Might Wind Up on the Menu’ — “Also during his talk, Nash stressed the position that ‘copyright is not the enemy of innovation’ and that ‘market-based solutions are the answer’ to the challenges AI poses to intellectual property-based industries like music.”

EU’s AI code of practice for companies to focus on copyright, safety — “The code is part of the AI rule book, which will come into effect in a staggered manner and will apply to Google owner Alphabet, Facebook owner Meta, OpenAI, Anthropic, Mistral and other companies. Signatories to the code will have to draw up and make publicly available summaries about the content used to train their general-purpose AI models, only employ copyright-protected content when using web crawlers as well as mitigate the risk of copyright-infringing output.”

Pirate IPTV Trio Sentenced to 14 Years Prison For Money Laundering — “A court in Brazil has handed down prison sentences totaling 14 years to three individuals for money laundering offenses linked to a pirate IPTV service. A software developer who made the service available through a popular IPTV app, his sister, and a mutual friend handled thousands of transactions through bank accounts, with funds laundered through a small hosting company.”

Use of Floorplans in Real Estate Listings is Permissible Under U.S. Copyright Law — “Initially, the district court ruled that the floorplan drawings fell within an exception of the Copyright Act that exempts pictures, paintings, photographs or other pictorial representations of an architectural work if it is visible from a public place. The appeals court rejected that theory and remanded the case for further development, observing that a fair use defense might apply. The case found its way back to the Eighth Circuit after the district court granted summary judgment on the basis of the defendants’ fair use.”

Bartz, Kadrey, and the Market for AI Training

By , July 10, 2025.

The first two court decisions (Bartz v. Anthropic and Kadrey v. Meta) addressing the use of copyrighted works to train generative AI were released in late June in rapid fire succession. Both came from judges within the same federal district, and both ruled, based on the records before them, that the use was fair. But despite the same outcomes, there were significant differences in how they got there.

Given the fact that these decisions may eventually be appealed, and the many lawsuits pending in multiple other US jurisdictions, this is far from the final word on this issue, and I suspect it will be a while before any judicial consensus begins to emerge.

But it is worth noting there are at least two points on which both district courts agree: (1) that the use of copyrighted works to train an AI models is “transformative”, and (2) courts should ignore harm to the market for licensing works under the fourth fair use factor. I previously wrote about why I think the courts are wrong on the first point, and today I want to write about why I think they are wrong on the second point.

Defining market harm

The fourth fair use factor, “undoubtedly the single most important element of fair use,”1Harper & Row, Publrs. v. Nation Enters., 471 U.S. 539, 566 (1985); see also Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 574 (1994) (quoting Harper & Row); Google LLC v. Oracle Am., Inc., 593 U.S. 1, 50 n.5 (2021) (Thomas, J., dissenting) (same). directs courts to consider “the effect of the use upon the potential market for or value of the copyrighted work.”217 U.S.C. § 107. It involves looking at both harm to the original (i.e., does the new work serve as a substitute for the original work) and harm to the market for derivative works.3471 U.S. at 568. This is not a damages analysis in the sense of focusing only on a measurement of already occurred harm.4Sony Corp. of Am. v. Universal City Studios, Inc., 464 U.S. 417, 451 (1984) (“Actual present harm need not be shown; such a requirement would leave the copyright holder with no defense against predictable damage.”). It is instead meant to focus prospectively, and look at not just the harm from the “particular actions of the alleged infringer,” but also the harm if the conduct would become widespread.5471 U.S. at 568 (“to negate fair use one need only show that if the challenged use ‘should become widespread, it would adversely affect the potential market for the copyrighted work.'”).

The plaintiffs in Bartz and Kadrey both advanced multiple forms of market harm caused by defendants resulting from their copying of works to train their AI models. Both provided evidence of a market for licensing copyrighted works for AI training.6Decl. of James Malackowski at 15-26, Bartz v. Anthropic, No. 3:24-cv-05417-WHA (N.D. Cali. April 25, 2025), ECF No. 156; Pl.’s Reply to Mot. for Partial Summ. J. and Opp’n to Meta’s Mot. for Partial Summ. J. at 26-30, Kadrey v. Meta, No. 3:23-cv-03417-VC (N.D. Cali. April 7, 2025), ECF No. 517. But in both cases, the courts dismissed this evidence as legally irrelevant, a conclusion that was a shock to many, especially given that this market is estimated to be worth $2.5 billion today and expected to grow to $30 billion within the decade.

Bartz provided little explanation for its conclusion, saying only,

A market [for licensing copyrighted works for training] could develop. Even so, such a market for that use is not one the Copyright Act entitles Authors to exploit. None of the cases cited by Authors requires a different result. All contemplated losses of something the Copyright Act properly protected — not the kinds of fair uses for which a copyright owner cannot rightly expect to control. 

Kadrey offered somewhat more explanation, saying,

But whether such a market [for licensing works to train AI] exists or is likely to develop is irrelevant, because this market is not one that the plaintiffs are legally entitled to monopolize. In every fair use case, the “plaintiff suffers a loss of a potential market if that potential [market] is defined as the theoretical market for licensing” the use at issue in the case. Therefore, to prevent the fourth factor analysis from becoming circular and favoring the rightsholder in every case, harm from the loss of fees paid to license a work for a transformative purpose is not cognizable.

In short, both courts decided copyright owners are not entitled to the market for licensing their works to AI developers for training, and thus courts should not consider that market when engaging in a fair use analysis.

No doubt, lawyers in these and other AI cases will argue that both courts seriously misread the case law. For example, it is notable that neither judge cites to Campbell v. Acuff-Rose, the one Supreme Court decision that spoke directly to what markets a copyright owner is entitled to under the fourth fair use factor, and how it seems unreasonable to twist Campbell’s reluctance to infer harm to a potential licensing market that copyright owners are unlikely to develop into a rule that courts should ignore an actual licensing market that copyright owners are actively engaged in.7510 U.S. 569, 592 (1994). They may also point out that both courts have engaged in the type of categorical, bright-line rules that are disfavored in fair use. They have impermissibly collapsed the fair use analysis into a single inquiry of transformativeness, only a few years after Warhol warned otherwise.8Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 550 (2023) (“The Court has cautioned that the four statutory fair use factors may not ‘be treated in isolation, one from another. All are to be explored, and the results weighed together, in light of the purposes of copyright.'”). There are likely additional legal arguments that would persuade an appellate court to reverse these holdings.

When markets advance the goals of copyright

But it is also worth looking at this issue through the lens of whether recognition of a market under the fourth factor would advance the goals of copyright. That is, after all, the basic premise of copyright.9471 U.S. at 558 (“By establishing a marketable right to the use of one’s expression, copyright supplies the economic incentive to create and disseminate ideas”); Eldred v. Ashcroft, 537 U.S. 186, 212 n.18 (2003) (“The economic philosophy behind the Copyright Clause is the conviction that encouragement of individual effort by personal gain is the best way to advance public welfare through the talents of authors and inventors. Accordingly, copyright law celebrates the profit motive, recognizing that the incentive to profit from the exploitation of copyrights will redound to the public benefit by resulting in the proliferation of knowledge. The profit motive is the engine that ensures the progress of science (cleaned up).”). Each of the four fair use factors must be considered independently in light of this goal. If licensing would serve the purpose of copyright better than unpermissioned and uncompensated use, shouldn’t that weigh against fair use?

It’s long been observed that a licensing market for criticism and parody would do little to promote the goals of copyright.10See, e.g., Wendy Gordon, Fair Use as Market Failure: A Structural and Economic Analysis of the Betamax Case and its Predecessors, 82 Colum. L. Rev. 1600, 1632-1634 (1982). Recognizing such a market might, for example, distort how critical uses are produced, which in turn would distort discourse in socially undesirable ways.

By contrast, a licensing market for AI training would promote the goals of copyright in at least four ways:

First, licensing creates incentives for high-quality data. A lot of work is needed to curate training datasets for specific purposes, clean and prepare the data to optimize model performance, and translate it into training-ready formats. With a market for licensing training datasets, publishers and other copyright owners will be motivated to make their own catalogs of works more attractive to AI developers. Third party aggregators and platforms will also be incentivized to curate and provide high-quality datasets with a competitive marketplace. In the absence of a market, only the largest companies will have the time and money to do this work, and they will be disincentivized from sharing datasets to prevent competitors from free-riding off their investments, inhibiting the broadest dissemination of copyrighted works for training.

Second, licensing reinforces public access. If scraping publicly available content is allowed without compensation, publishers are given a clear incentive to move their works behind paywalls, authentication barriers, or technical protections. Ironically, in the name of free access for AI, fair use may end up reducing access for humans. But if publishers know that public access does not mean uncompensated use, they are more likely to keep their content openly available, confident that doing so will not render it a free resource for commercial appropriation.

Third, licensing is pro-competitive. It lowers barriers to entry and broadens access to high-quality datasets, which means more AI companies can compete, leading to increased innovation, product variety, and lower costs. Licensing enables specialization, allowing licensors to deepen their expertise and rewarding them in the market for increasing efficiencies in creating and providing training datasets. Contrary to claims that licensing would favor big companies over small, it is the world that ignores copyright that is a winner takes all world, where the competitive edge goes only to the largest companies that can devote the most resources to collecting and building training datasets.

Fourth, licensing creates certainty. “Because copyright law ultimately serves the purpose of enriching the general public through access to creative works,” says the Supreme Court, “it is peculiarly important that the boundaries of copyright law be demarcated as clearly as possible.”11Fogerty v. Fantasy, Inc., 510 U.S. 517, 527 (1994). The use of copyrighted works by AI developers is complex, far more complex than uses like criticism or parody,12Though even a parodic work “may have a more complex character,” as the Supreme Court noted in Campbell, “with effects not only in the arena of criticism but also in protectible markets for derivative works, too.” 510 U.S. at 592. with multiple potential uses on the ingestion side and potential reproductions occurring on the output side. Even if courts begin to find that the actual training with copyrighted works is fair use, that still leaves substantial legal uncertainty around different factual scenarios, ancillary uses of works, and downstream uses.

The decision in Bartz is proof positive of this. Although the court held that the copies of works used to train Anthropic’s LLM were justified under fair use, it concluded that the downloading of pirate copies was not fair use, and any internal copying that was not part of the training process may still be infringing. Licensing can prospectively address these issues with far greater certainty and allow parties to allocate risks more efficiently, leading to increased investment and innovation.

Perhaps most importantly when it comes to the constitutional goals of copyright: licensing rewards authors and publishers.13Kirtsaeng v. John Wiley & Sons, Inc., 579 U.S. 197, 204 (2016) (“copyright law ultimately serves the purpose of enriching the general public through access to creative works. The statute achieves that end by striking a balance between two subsidiary aims: encouraging and rewarding authors’ creations while also enabling others to build on that work.”) (citing Fogerty, 510 U. S., at 526). Creators of books, journalism, photography, and other expressive works are essential to the cultural and informational commons. Licensing allows them to share in the commercial success of AI, providing income streams that support continued creation. The public continues to benefit from their creative work, and the next generation of AI benefits from a renewable source of high-quality training materials.

Conclusion

Both Bartz and Kadrey have created a bright-line rule that transformativeness equals fair use. A categorical rule that excludes consideration of harm to markets for uses that a court determines are transformative effectively collapses the four factor analysis into a single inquiry. If this rule stands, it would put fair use at odds with the goals of copyright, and turn it into a doctrine that exists for the private benefit of the largest commercial actors in the world at the expense of the public good.

References

References
↑1 Harper & Row, Publrs. v. Nation Enters., 471 U.S. 539, 566 (1985); see also Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 574 (1994) (quoting Harper & Row); Google LLC v. Oracle Am., Inc., 593 U.S. 1, 50 n.5 (2021) (Thomas, J., dissenting) (same).
↑2 17 U.S.C. § 107.
↑3 471 U.S. at 568.
↑4 Sony Corp. of Am. v. Universal City Studios, Inc., 464 U.S. 417, 451 (1984) (“Actual present harm need not be shown; such a requirement would leave the copyright holder with no defense against predictable damage.”).
↑5 471 U.S. at 568 (“to negate fair use one need only show that if the challenged use ‘should become widespread, it would adversely affect the potential market for the copyrighted work.'”).
↑6 Decl. of James Malackowski at 15-26, Bartz v. Anthropic, No. 3:24-cv-05417-WHA (N.D. Cali. April 25, 2025), ECF No. 156; Pl.’s Reply to Mot. for Partial Summ. J. and Opp’n to Meta’s Mot. for Partial Summ. J. at 26-30, Kadrey v. Meta, No. 3:23-cv-03417-VC (N.D. Cali. April 7, 2025), ECF No. 517.
↑7 510 U.S. 569, 592 (1994).
↑8 Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 550 (2023) (“The Court has cautioned that the four statutory fair use factors may not ‘be treated in isolation, one from another. All are to be explored, and the results weighed together, in light of the purposes of copyright.'”).
↑9 471 U.S. at 558 (“By establishing a marketable right to the use of one’s expression, copyright supplies the economic incentive to create and disseminate ideas”); Eldred v. Ashcroft, 537 U.S. 186, 212 n.18 (2003) (“The economic philosophy behind the Copyright Clause is the conviction that encouragement of individual effort by personal gain is the best way to advance public welfare through the talents of authors and inventors. Accordingly, copyright law celebrates the profit motive, recognizing that the incentive to profit from the exploitation of copyrights will redound to the public benefit by resulting in the proliferation of knowledge. The profit motive is the engine that ensures the progress of science (cleaned up).”).
↑10 See, e.g., Wendy Gordon, Fair Use as Market Failure: A Structural and Economic Analysis of the Betamax Case and its Predecessors, 82 Colum. L. Rev. 1600, 1632-1634 (1982).
↑11 Fogerty v. Fantasy, Inc., 510 U.S. 517, 527 (1994).
↑12 Though even a parodic work “may have a more complex character,” as the Supreme Court noted in Campbell, “with effects not only in the arena of criticism but also in protectible markets for derivative works, too.” 510 U.S. at 592.
↑13 Kirtsaeng v. John Wiley & Sons, Inc., 579 U.S. 197, 204 (2016) (“copyright law ultimately serves the purpose of enriching the general public through access to creative works. The statute achieves that end by striking a balance between two subsidiary aims: encouraging and rewarding authors’ creations while also enabling others to build on that work.”) (citing Fogerty, 510 U. S., at 526).

Friday’s Endnotes – 06/27/25

By , June 27, 2025.

Fair Use Decision Fumbles Training Analysis but Sends Clear Piracy Message — “First analyzing Anthropic’s use of copyrighted works to train its Claude large language model (LLM), Judge William Alsup immediately falls into a fair use trap that the Supreme Court warned against in its seminal 2023 Warhol v. Goldsmith decision, letting his foregone conclusion about the transformative nature of the use control the rest of the fair use analysis.”

U.S. Copyright Office Replaces Online Public Catalog with Copyright Public Records System — “Today, the U.S. Copyright Office is pleased to announce that the Copyright Public Records System (CPRS) has replaced its Online Public Catalog. CPRS provides copyright registration and recordation data with advanced search capabilities, filters, and improved interfaces for public users and Office staff. … CPRS includes both recordation and registration information from 1978 to the present and searchable metadata for over 3.8 million registration applications from 1898 to 1945.”

Unlocking Creativity: The Socioeconomic Benefits of Copyright — “In a world increasingly driven by ideas, innovation, and digital transformation, copyright stands as a cornerstone of economic vitality and cultural expression. This report is a comprehensive literature review that explores the powerful role copyright plays in fueling creativity, supporting high-quality jobs, and driving global competitiveness through creative sectors.”

Getty Images Drops Main Copyright Claims Against Stability AI in UK Legal Case — “The withdrawn claims centered on Getty’s assertion that Stability trained Stable Diffusion using millions of copyrighted images without authorization, including some that contained Getty’s distinct watermark. However, legal experts say the company likely struggled to establish that any alleged infringement occurred under U.K. jurisdiction.”

Denmark to tackle deepfakes by giving people copyright to their own features — “The Danish government said on Thursday it would strengthen protection against digital imitations of people’s identities with what it believes to be the first law of its kind in Europe.”

Not That Kind of Transformation: AI, Fair Use, and Relating Back

By , June 25, 2025.

In Bartz v. Anthropic, Judge Alsup found the use of copyrighted works to train a large language model to be “justified as a fair use”, describing the technology as “among the most transformative many of us will see in our lifetimes.”

Generative AI may certainly turn out to be remarkably transformative in the ordinary sense of the word. But, contrary to the legal conclusion reached by Judge Alsup, I argue that the use of copyrighted works for training doesn’t constitute “transformative” use in the context of copyright law.

In this context, a use is not transformative merely because it produces something new or technologically sophisticated. Rather, as established through decades of judicial interpretation, a transformative use is one that relates back to the original work by creating new information and insights about that work. Generative AI does not do this. It instead reappropriates the expressive content of the work to enable the generation of synthetic expressive content completely divorced from the original work.

This is not a criticism of the technology itself. Generative AI may well be impressive, even revolutionary. But in its recent recalibration of transformativeness, the Supreme Court was clear that not every innovative or creative secondary use should be considered transformative.1Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 543 n.18 (2023) (“The [Google] Court did not hold that any secondary use that is innovative, in some sense, or that a judge or Justice considers to be creative progress consistent with the constitutional objective of copyright, is thereby transformative.”).

Justification

The 2023 Warhol Foundation v. Goldsmith decision marked the first time in nearly 30 years that the Supreme Court directly addressed transformativeness, and the period in between saw a steady expansion of the doctrine, which some worried went too far.2Kienitz v. Sconnie Nation LLC, 766 F.3d 756, 758 (7th Cir. 2014) (“We’re skeptical of Cariou‘s approach, because asking exclusively whether something is ‘transformative’ not only replaces the list in §107 but also could override 17 U.S.C. §106(2), which protects derivative works.”; TCA TV Corp. v. McCollum, 839 F.3d 168, 181 (2d Cir. 2016) (“Insofar as Cariou might be thought to represent the high-water mark of our court’s recognition of transformative works, it has drawn some criticism… We need not defend Cariou here.”).

Warhol regained control of the doctrine, trimmed it of its excesses and refocused it. The Court warned against looking at merely whether something new is created—“Most copying has some further purpose… Many secondary works add something new.”3598 U.S. at 528. The key, Warhol reminds, is to look at the justification for the use. This involves two senses of the word. First, a use is justified if it is the type of use that furthers the purpose of copyright without prejudicing the original author. And second, the use is justified if the user needs the original work to serve this purpose.

Justification in the first sense looks toward the purpose of copyright itself, which is “the conviction that encouragement of individual effort by personal gain is the best way to advance public welfare through the talents of authors and inventors in ‘Science and useful Arts.’”4Mazer v. Stein, 347 U.S. 201, 219 (1954). Exclusive rights allow creators to control and monetize their works. Markets built around those rights reflect diverse preferences—economic, aesthetic, educational, scientific—and empower authors and publishers to pursue their own goals.

Fair use is an exception to these exclusive rights, but it is an exception intended to serve the same purpose as copyright overall. Courts must therefore exercise care when departing from the general rule of exclusive rights, since “underprotection of copyright disserves the goals of copyright just as much as overprotection.”5Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 599 (1994) (J. Kennedy, concurring).

Relating Back

The hallmark of justification, and thus transformativeness, is that the new use “relates back” to the original. The public benefit is two-fold for such uses: the public benefits from the new use itself, and it benefits from the insights about the original work that are created.

This relation-back principle is reflected in the statute, which identifies illustrative purposes—“criticism, comment, news reporting, teaching…, scholarship, or research” all or which involve new uses that provide new information or insights about the original work or treat them as referential objects in the context of further discussion.617 U.S.C. § 107.

The principle is also at the core of Campbell’s parody/satire distinction, which drew the line between a work that “at least in part, comments on” the original work being parodied (parody) and a work where “the commentary has no critical bearing on the substance or style of the original composition” and the original work is used merely “to get attention or to avoid the drudgery in working up something fresh” (satire).7510 U.S. at 580.

And the legislative history recalls this relating-back characteristic. In the Copyright Office’s preliminary study on fair use that kicked off the drafting effort for the 1976 Copyright Act, Alan Latman wrote,

The modus operandi of certain fields requires that the rights of each author yield to a step-by-step progress. This consideration is often linked to the constitutional support for fair use as an indispensable tool in the promotion of “science.” Practical necessity and constitutional desirability are strongest in the area of scholarly works.

Similarly, in reviews of a work, a certain amount of reconstruction is often necessary; and in burlesque, the user must be permitted to accomplish the “recalling or conjuring up of the original.”

As a corollary, this relating-back purpose would be stymied without the ability to use the original work; because the new use is tied to the original work, there are no substitutes for the original work. Thus, we see courts reject transformativeness when the original work is used as a mere commodity or is fungible to the ultimate purpose.8839 F.3d at 182 (“The ‘dramatic’ purpose served by the Routine in the Play appears to be as a ‘McGuffin,’ that is, as a theatrical device that sets up the plot, but is of little or no significance in itself. To advance the plot of the Play, specifically, to have the puppet Tyrone take on a persona distinct from that of Jason, defendants needed Jason to lie about something and for Tyrone to call him on it. But the particular subject of the lie—the Routine—appears irrelevant to that purpose.”); Monge v. Maya Magazines, Inc., 688 F.3d 1164, 1175 (9th Cir. 2012) (“the controversy here has little to do with photos; instead, the photos here depict the couple’s clandestine wedding. The photos were not even necessary to prove that controverted fact—the marriage certificate, which is a matter of public record, may have sufficed to inform the public that the couple kept their marriage a secret for two years.”).

The principle holds true even when looking at technological uses. For example, like criticism and commentary, technological tools that create new information about existing works, like book and image search tools, have been found transformative.9Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015) (book search); Authors Guild, Inc. v. HathiTrust, 755 F.3d 87 (2d Cir. 2014) (book search); Perfect 10, Inc. v. Amazon.com, Inc., 508 F.3d 1146 (9th Cir. 2007) (image search); Kelly v. Arriba Soft Corp., 336 F.3d 811 (9th Cir. 2003) (image search). Second, as with news reporting and biographical uses, technological tools that analyze specific works, like a plagiarism detector, treat the works as discrete referential objects to provide increased understanding of those specific works.10A.V. v. iParadigms, 562 F.3d 630 (4th Cir. 2009). In each of these cases, the technological use relates back to the original work that was copied.

Training Does Not Relate Back

That is not true in the case of training a generative AI model.

Training a large language model itself starts by breaking down huge amounts of text into smaller parts, and converting those parts into numbers for processing. These numbers (called tokens) are arranged in a mathematical space that captures how often tokens appear together and their contextual relationships. In other words, the unique choices each writer makes, the specific words they select and how they arrange them, are compiled into a single mathematical representation that encodes patterns of usage and co-occurrence.

The model then uses this representation to create a function optimized to predict the most likely next token in a sequence. It does not understand or think like a person but generates what appears as coherent and relevant text by operating on statistical patterns it has derived from the training materials.

This is not transformative in the legal sense.

Crucially, the training process does not comment on or critique the training materials. It does not relate back to them in any meaningful or referential way. What is taken is a part of, not information about, the original authors’ creative expression.11Accord Video Pipeline, Inc. v. Buena Vista Home Entm’t, Inc., 342 F.3d 191, 198 n.5 (3d Cir. 2003) (“We note that the clip previews do not constitute mere ‘information’ about the movies, as would, for example, a list of the names of the actors starring in a film, or a statement of the rating it received… Regardless, the clips are part of — not information about — Disney’s expressive creations.”). The original expressive choices are extracted and embedded into the model’s internal representation, not to analyze them, but to reproduce similarly structured expression.

Indeed, in many cases, the process does not even identify which training materials were most influential in producing a given output. The process is opaque by design. In that sense, the AI model is the opposite of a referential or critical use. It is a black box whose output conceals rather than illuminates its sources.

Because there is no relating back, then the use of a particular work is also not justified in the narrow sense of the term. The individual works used in training are entirely fungible, and developers have a universe of available substitutes that could achieve the same purpose. The public benefit is served best in such circumstances through the ordinary application of exclusive rights in the market.

Conclusion

Generative AI models may “transform” input data into dazzling new outputs, but this is not the transformation that fair use favors. Under fair use, a transformative use is one that comments on, critiques, or provides new insights about an original work. It is not simply a technological process that digests and reuses expressive content in a different form.

Generative AI training reappropriates the expressive elements of copyrighted works to enable the generation of new content. It does not point back to, analyze, or even acknowledge the originals. That makes it more like satire than parody—creative, yes, but not legally justified without permission.

The choice to allow one’s work to be ingested into a generative system—to become raw material for future outputs—belongs to the copyright owner. Courts should be careful not to let technological innovation obscure that basic principle of copyright law and undermine its ability to benefit the public.

References

References
↑1 Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 543 n.18 (2023) (“The [Google] Court did not hold that any secondary use that is innovative, in some sense, or that a judge or Justice considers to be creative progress consistent with the constitutional objective of copyright, is thereby transformative.”).
↑2 Kienitz v. Sconnie Nation LLC, 766 F.3d 756, 758 (7th Cir. 2014) (“We’re skeptical of Cariou‘s approach, because asking exclusively whether something is ‘transformative’ not only replaces the list in §107 but also could override 17 U.S.C. §106(2), which protects derivative works.”; TCA TV Corp. v. McCollum, 839 F.3d 168, 181 (2d Cir. 2016) (“Insofar as Cariou might be thought to represent the high-water mark of our court’s recognition of transformative works, it has drawn some criticism… We need not defend Cariou here.”).
↑3 598 U.S. at 528.
↑4 Mazer v. Stein, 347 U.S. 201, 219 (1954).
↑5 Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 599 (1994) (J. Kennedy, concurring).
↑6 17 U.S.C. § 107.
↑7 510 U.S. at 580.
↑8 839 F.3d at 182 (“The ‘dramatic’ purpose served by the Routine in the Play appears to be as a ‘McGuffin,’ that is, as a theatrical device that sets up the plot, but is of little or no significance in itself. To advance the plot of the Play, specifically, to have the puppet Tyrone take on a persona distinct from that of Jason, defendants needed Jason to lie about something and for Tyrone to call him on it. But the particular subject of the lie—the Routine—appears irrelevant to that purpose.”); Monge v. Maya Magazines, Inc., 688 F.3d 1164, 1175 (9th Cir. 2012) (“the controversy here has little to do with photos; instead, the photos here depict the couple’s clandestine wedding. The photos were not even necessary to prove that controverted fact—the marriage certificate, which is a matter of public record, may have sufficed to inform the public that the couple kept their marriage a secret for two years.”).
↑9 Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015) (book search); Authors Guild, Inc. v. HathiTrust, 755 F.3d 87 (2d Cir. 2014) (book search); Perfect 10, Inc. v. Amazon.com, Inc., 508 F.3d 1146 (9th Cir. 2007) (image search); Kelly v. Arriba Soft Corp., 336 F.3d 811 (9th Cir. 2003) (image search).
↑10 A.V. v. iParadigms, 562 F.3d 630 (4th Cir. 2009).
↑11 Accord Video Pipeline, Inc. v. Buena Vista Home Entm’t, Inc., 342 F.3d 191, 198 n.5 (3d Cir. 2003) (“We note that the clip previews do not constitute mere ‘information’ about the movies, as would, for example, a list of the names of the actors starring in a film, or a statement of the rating it received… Regardless, the clips are part of — not information about — Disney’s expressive creations.”).

Friday’s Endnotes – 06/20/25

By , June 20, 2025.

Study: Meta AI model can reproduce almost half of Harry Potter book — “These results give everyone in the AI copyright debate something to latch onto. For AI industry critics, the big takeaway is that—at least for some models and some books—memorization is not a fringe phenomenon.”

Bots are overwhelming websites with their hunger for AI data — “‘The cultural institutions that host online collections are not resourced to continue adding more servers, deploying more sophisticated firewalls, and hiring more operations engineers in perpetuity,’ the report says. ‘That means it is in the long-term interest of the entities swarming them with bots to find a sustainable way to access the data they are so hungry for.'”

Supreme Court declines to hear three IP cases — “Copyright discovery rule. The Court declined to hear the petition in RADesign, Inc. v. Michael Grecco Productions, Inc., Dkt No. 24-1137, letting stand a ruling by the U.S. Court of Appeals for the Second Circuit allowing a photographer’s copyright infringement case against a shoe designer to proceed.”

Beijing court hands AI copyright violators up to 18 months in prison — “The four defendants were accused of using AI software to alter original illustrations found online, creating more than 3,000 jigsaw puzzles for sale and making illegal profits exceeding 270,000 yuan ($37,556), prosecutors said. The court found them guilty under a criminal indictment from district prosecutors, marking what is reportedly the first criminal ruling in Beijing involving copyright infringement through generative AI.”

Music Publishers and X Begin ‘Good Faith’ Negotiations in Copyright Lawsuit — “’Twitter [now X] stands alone as the largest social media platform that has completely refused to license the millions of songs on its service,’ National Music Publishers Association president and CEO David Israelite said in a statement around the time of the suit’s filing. ‘Twitter knows full well that music is leaked, launched, and streamed by billions of people every day on its platform.’”

Friday’s Endnotes – 06/13/25

By , June 13, 2025.

Say ‘No’ to Unlicensed AI Training! — The Copyright Alliance kicked off a grassroots AI campaign that encourages the public to send letters to Members of Congress and President Trump in anticipation of the administration releasing its AI Action Plan in mid-July 2025. The letters urge elected officials to protect the rights and livelihoods of creators by rejecting laws and policies that would allow AI companies to train on their work without consent and compensation. Those interested in participating may use the letters “as is,” revise them, or write their own.

London AI firm says Getty copyright case poses ‘overt threat’ to industry — “Getty’s case against Stability AI for copyright and trademark infringement relating to its vast photography archives reached the high court in London on Monday.” Bonus content—IPKitten has a series of posts covering the legal issues raised in the lawsuit: UK trial begins… (Part 1), Part 2 – copyright and database right, Part 3 – Defences.

In first-of-its-kind lawsuit, Hollywood giants sue AI firm for copyright infringement — “Many companies have gone after AI firms for copyright infringement, such as The New York Times (which sued OpenAI and Microsoft), Sony Music Entertainment (which filed a suit against AI song generator startups Suno and Udio) and Getty Images (against Stability AI). But this is the first time major Hollywood players have joined the fight against the AI landscape.”

AI-driven system helps cut copyright disputes in Textile City — “China’s Textile City, the world’s largest textile distribution hub located in Keqiao, handles about a quarter of the global fabric trade and exports textile products to more than 190 countries and regions. The bustling business also sees frequent disputes related to fabric patterns, said Li Zisu, vice-president of the Keqiao District People’s Court. She revealed that between 2008 and 2020, more than 5,000 merchants were involved in pattern-related lawsuits, and said that the biggest challenge in handling these cases was determining ownership of the pattern copyright.”

Italy probes Meta over music copyright negotiations — “The Italian Competition Authority said it was probing Meta’s alleged abuse of economic dependence of the Italian Society of Authors and Publishers (SIAE), the public authority charged with protecting artists’ copyright in Italy. The SIAE had a contract with Meta that expired in December 2022.”

Friday’s Endnotes – 05/30/25

By , May 30, 2025.

Getty Images spending millions to battle a ‘world of rhetoric’ in AI suit, CEO says — “Technology startups like OpenAI, Anthropic and Mistral have flourished by taking vast amounts of data from the open web and using it to train their foundational AI models, which can produce lifelike texts, images and videos. However, the strategies of these firms have raised concerns over their use of copyrighted material. Several lawsuits have targeted AI firms over alleged copyright infringements from The New York Times’ suit against OpenAI to several U.S. record labels’ claims against AI music generation services Suno and Udio.”

Generative AI’s Illusory Case for Fair Use — “Despite wide employment of anthropomorphic terms to describe their behavior, AI machines do not learn or reason as humans do. Instead, they employ an algorithmic process to store the works they are fed during the training process. They do not ‘know’ anything independently of the works on which they are trained, so their output is a function of the copied materials… The exploitation of expressive content to produce new expressive content sharply distinguishes AI copying from the copying at issue in the technological fair use cases relied upon by AI’s fair use advocates.”

Car in Gone in Sixty Seconds is not entitled to copyright protection — “Prior Ninth Circuit precedent, DC Comics v. Towle, 802 F.3d 1012 (9th Cir. 2015), establishes a test to determine whether a character is entitled to copyright protection: (1) the character must have ‘physical as well as conceptual qualities,’ (2) the character must be ‘sufficiently delineated to be recognizable as the same character whenever it appears’ and display ‘consistent, identifiable character traits and attributes,’ and (3) the character must be ‘especially distinctive’ and contain ‘some unique elements of expression.’ The Eleanor character failed all three parts of the test, the court of appeals found.”

Judge Rejects Ex-Copyright Chief’s Bid to Pause Trump Firing — “Even if Perlmutter was likely to succeed on the merits, [Judge] Kelly said his analysis was pinned on the irreparable harm threshold required to grant the order and Perlmutter did not meet that requirement. Perlmutter’s argument of irreparable harm rested on the idea that she was deprived of the statutory right to function as the register of copyrights, Kelly said. Recent court decisions from the US Supreme Court and the D.C. Circuit have been ‘skeptical’ of that argument, the judge said.”

The ‘beige Amazon influencer’ lawsuit is headed for dismissal — “The lawsuit was simultaneously disconcerting and benign, eerie and borderline comical: the story of two women whose lives had begun to resemble each other’s via social media platforms made for a compelling storyline. The cream, white, and beige aesthetic of their content (and lives) meant that the essence of what was allegedly infringed was commonplace, even basic — but the similarities, documented over dozens of examples submitted to the court, were strange nonetheless”

Generative AI, Copyright, and Market Harm

By , May 28, 2025.

Many of the leading artificial intelligence developers have trained their generative AI models on copyrighted materials without the permission of the copyright owners. As courts and the Copyright Office have examined whether such use should be allowed under fair use, some have considered the fact that these tools can quickly generate works that may not be similar in the legal copyright sense to the training works, but nevertheless compete with the original works because they are close substitutes. Recognizing this harm under fair use is not, as some have responded, an improper expansion of the scope of copyright protection. Rather, it is a reasonable calibration of a purposefully flexible doctrine to a novel technology.

Kadrey v. Meta

Earlier this month, lawyers for a group of authors squared off against lawyers for Meta in front of Judge Chhabria in a federal district court in California to argue about whether the use of copyrighted works without permission from the copyright owner to train a generative AI models should be allowed under the law. There have been over 40 lawsuits filed since January 2023 from copyright owners against various AI developers over this issue. Kadrey v. Meta is the first one to reach a hearing on the merits of the copyright and fair use issues.

Fair use, of course, requires courts to consider and balance four factors to determine whether the use of a copyrighted work that would otherwise be infringing should be permitted. The fourth factor, “the effect of the use upon the potential market for or value of the copyrighted work”, is the most important factor,1 Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 555 (2023); Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 574 (1994); Harper & Row, Publrs. v. Nation Enters., 471 U.S. 539, 566 (1985) (“This last factor is undoubtedly the single most important element of fair use”). and it typically involves courts looking at whether the new use would usurp sales or licenses of the original work, or undermine markets for derivative works.

During the hearing Judge Chhabria carefully and thoughtfully prodded the lawyers on their arguments. While questioning Meta’s attorney, Chhabria focused in on the ability of generative AI models to quickly produce coherent and humanlike textual material, a function that courts have not yet considered in a fair use case—and he asked why the impact of this feature should not be considered under the fourth fair use factor.

“You have companies using copyright-protected material to create a product that is capable of producing an infinite number of competing products,” said Chhabria. “You are dramatically changing, you might even say obliterating, the market for that person’s work, and you’re saying that you don’t even have to pay a license to that person. I just don’t understand how that can be fair use.”2Blake Brittain, Judge in Meta Case Warns AI Could ‘Obliterate’ Market for Original Works, Reuters (May 1, 2025), https://www.reuters.com/legal/litigation/judge-meta-case-weighs-key-question-ai-copyright-lawsuits-2025-05-01/.

Copyright Office

Shortly afterward, the U.S. Copyright Office issued a pre-publication version of its report on copyright and artificial intelligence, covering generative AI training. Like Judge Chhabria, the Office considered the impact of generative AI outputs generally on the market of individual copyrighted works. It concluded, among other things, that courts should consider the harms caused where a generative AI model’s outputs, even if not substantially similar to a specific copyrighted work, compete in the market for that type of work. The Office rejected calls to read the fourth fair use factor too narrowly.

The statute on its face encompasses any “effect” upon the potential market. The speed and scale at which AI systems generate content pose a serious risk of diluting markets for works of the same kind as in their training data. That means more competition for sales of an author’s works and more difficulty for audiences in finding them. If thousands of AI-generated romance novels are put on the market, fewer of the human-authored romance novels that the AI was trained on are likely to be sold. Royalty pools can also be diluted.

“Market dilution”

Professor Pam Samuelson has said, “Congress expected the fair use doctrine to evolve when it passed the 1976 Act, and evolve it certainly has.”3Pamela Samuelson, Possible Futures of Fair Use, 90 Wash. L. Rev. 815, 863-64 (2015). It makes sense that the doctrine’s evolution would continue when confronted with the use of copyrighted works to train generative AI models, a use that is not closely analogous with uses in existing fair use precedent.

Almost immediately after release of the Copyright Office report, there was a strand of criticism of the Office’s “market dilution” conclusion that, as I discuss below, misconstrues (inadvertantly, perhaps) the fair use framework. The result is an argument that may seem superficially compelling but upon closer examination is a logical fallacy.

The following are presented as illustrative examples of this argument:

  • Granting copyright holders the ability to stop AI’s generative function because it creates new, noninfringing works would expand the limited monopoly of copyright far beyond its permissible bounds. … copyright is limited to infringement of specific works—and does not give any right to copyright holders to monopolize methods for creation or to prevent general competition in the marketplace presented by noninfringing but competing works.4Edward Lee, Fair Use and the Origin of AI Training (February 01, 2025). Houston Law Review (forthcoming 2025).
  • The ‘market dilution’ theory is definitely novel — and inconsistent with what SCOTUS has held to be the proper scope of the copyright monopoly. An author cannot get a vague proprietary claim to the whole market for a particular type of work simply by producing individual works of that type.5Annemarie Bridy (@annemariebridy), Bluesky (May 21, 2025), https://bsky.app/profile/annemariebridy.bsky.social/post/3lot4u3ej2k2c.

Essentially, the argument states that allowing courts to consider market dilution under the fourth fair use factor is synonymous with extending an author’s exclusive rights over competing but noninfringing works.

This is a classic strawman argument. The rebuttals are correct in stating that copyright protection does not extend beyond infringement of the specific work. The flaw is conflating the scope of infringement with the fair use analysis.

The copyright owner’s complaint against AI developers is that their work was used without permission to train an AI model. Their prima facie claim is complete when it’s been shown that a developer has encroached on one of the exclusive rights protected by copyright, a showing that is easily established, given that the training process will involve at least one reproduction of a copyrighted work.6U.S. Copyright Office, Copyright and Artificial Intelligence: Part III – The Use of Copyrighted Works in Training AI Models 26 (Pre-Publication Version) (2024). That’s it. The only thing left at that point, absent an affirmative defense, is to determine remedies.

Fair use is an affirmative defense.7Andy Warhol Found. for the Visual Arts, Inc. v.
Goldsmith, 598 U.S. 508, 547 n.21 (2023); Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 590 (1994); Harper & Row Publishers, Inc. v. Nation Enters., 471 U.S. 539, 560 (1985). The burden of proof is on the defendant.8This is a fundamental principle that has been established since ancient times. See, e.g., Dig. 22.3.2 (Ulpian, Ad Edictum 69) (“Ei incumbit probatio qui dicit, non qui negat.”). As part of that defense, a defendant must show that “the effect of the use upon the potential market for or value of the copyrighted work” has not prejudiced the copyright owner.

In other words, in this context, the copyright owner is not asserting a claim that the scope of their copyright enables them to block competing but non-infringing works. They are asserting simply that an AI developer has copied their work for training, which constitutes infringement. It is the AI developer who is claiming that this use should nevertheless be excused despite the clear prejudice it will cause authors of the original works.

While novel, I think it’s reasonable to consider the impact that a tool that can generate a nearly limitless number of non-infringing but competing outputs will have on the market for the original work when analyzing whether to allow the use of the work without permission. I think the statute is even broader in this regard than the Copyright Office suggests in its discussion. The statute not only encompasses any effect on the potential market, it also encompasses any effect on the “value of the copyrighted work.” Both the markets for and value of works could be obliterated by generative AI outputs.

The consideration of “market dilution” under the fourth fair use factor is thus not an expansion of copyright’s scope but a careful calibration of the privilege to use copyrighted works without permission to ensure that the use does not prejudice the rights of creators and undermine the public interest.

References

References
↑1 Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 555 (2023); Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 574 (1994); Harper & Row, Publrs. v. Nation Enters., 471 U.S. 539, 566 (1985) (“This last factor is undoubtedly the single most important element of fair use”).
↑2 Blake Brittain, Judge in Meta Case Warns AI Could ‘Obliterate’ Market for Original Works, Reuters (May 1, 2025), https://www.reuters.com/legal/litigation/judge-meta-case-weighs-key-question-ai-copyright-lawsuits-2025-05-01/.
↑3 Pamela Samuelson, Possible Futures of Fair Use, 90 Wash. L. Rev. 815, 863-64 (2015).
↑4 Edward Lee, Fair Use and the Origin of AI Training (February 01, 2025). Houston Law Review (forthcoming 2025).
↑5 Annemarie Bridy (@annemariebridy), Bluesky (May 21, 2025), https://bsky.app/profile/annemariebridy.bsky.social/post/3lot4u3ej2k2c.
↑6 U.S. Copyright Office, Copyright and Artificial Intelligence: Part III – The Use of Copyrighted Works in Training AI Models 26 (Pre-Publication Version) (2024).
↑7 Andy Warhol Found. for the Visual Arts, Inc. v.
Goldsmith, 598 U.S. 508, 547 n.21 (2023); Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 590 (1994); Harper & Row Publishers, Inc. v. Nation Enters., 471 U.S. 539, 560 (1985).
↑8 This is a fundamental principle that has been established since ancient times. See, e.g., Dig. 22.3.2 (Ulpian, Ad Edictum 69) (“Ei incumbit probatio qui dicit, non qui negat.”).

Friday’s Endnotes – 05/23/25

By , May 23, 2025.

Judge Hints Anthropic’s AI Training on Books is Fair Use — “Anthropic’s use of notorious digital piracy websites raised concerns for Alsup. In response, the company’s counsel, Joseph Richard Farris of Arnold & Porter Kaye Scholer LLP, argued the Supreme Court has been skeptical whether bad faith has any effect on the fair use analysis. Alsup pushed back, saying ‘I have a hard time seeing that you can commit what is ordinarily a crime, but get exonerated because you end up using it for a transformative use.'”

Copyright Alliance CEO and Others Resign from ALI’s Copyright Restatement Project — “In a statement issued today, Copyright Alliance CEO Keith Kupferschmid announced that he, along with numerous other advisers and liaisons, have resigned from the American Law Institute’s (ALI) Restatement of Copyright Law (Restatement) project, effective immediately. The resignations follow those of four prominent law professors earlier this week.”

Top copyright official sues Trump over firing — “As head of the nation’s copyright office, Perlmutter would have overseen registering copyright claims and maintaining records about copyright ownership. The office functioned within the Library of Congress by providing the copyright deposits that make up a significant part of the library’s collections. The copyright office is also responsible for conducting studies and advising Congress on copyright issues, including regarding the growing generative artificial intelligence industry.”

Photographer Loses Copyright Lawsuit Claiming Lil Nas X Stole His Poses — “Last week, the Ninth Circuit Court of Appeals decided that the competing Instagram posts had only a handful of minor similarities, which were not enough to constitute copyright infringement. The court also concluded that the poses and themes featured in Woodland’s photos are not protected under copyright law.”

Does Human Learning equal Machine Learning? High Court of Delhi to rule on lawfulness of TDM for Machine Learning — “While dozens of US District Courts are currently grappling with the question of whether AI training with protected works constitutes fair use, the UK High Court is largely grappling with jurisdictional questions, and EU courts are mainly concerned with the modalities of rights reservations … it is now the High Court of Delhi’s turn. “

Friday’s Endnotes – 05/16/25

By , May 16, 2025.

US Copyright Office Releases Highly Anticipated Report on Generative AI Training – Here’s What It Actually Says — “The U.S. Copyright Office (USCO) recently released its highly anticipated Report on Generative AI Training (the third and final part in the USCO’s AI and copyright series) in a pre-publication format. At a time when headlines dominate, and despite the leadership changes underway at the USCO, the report represents a substantial body of work and analysis that explores how copyright law applies to the training of generative AI systems with a level of nuance that reflects the expertise of its drafters and a clear understanding of the importance of both the copyright and AI sectors to the broader innovation ecosystem.”

Leading Scholars Insist Their Names Be Removed from the ALI Restatement of Copyright Law — “Four luminaries of copyright law and scholarship submitted a letter to the American Law Institute (ALI) formally withdrawing their names as Advisers from the Restatement of Copyright Law, approval of which is set to be voted on next week. Professors Shyam Balganesh, Jane Ginsburg, and Peter Menell, along with attorney David Nimmer submitted the May 12 letter conveying strong disagreement with both the substance of the Restatement and the subterfuge in the process.”

AI-Powered News Piracy Site Blocked By ISPs After Court Sides With Publishers — “A joint investigation by Libération and Next revealed that at least 1,000 similar sites churn out infringing content in much the same way. In some cases, AI ‘hallucinations’ aren’t noticed by site operators or the public, resulting in bogus automated news being taken as fact, then cited as source material for articles published on Wikipedia.”

Anthropic expert accused of using AI-fabricated source in copyright case — “A federal judge in San Jose, California, on Tuesday ordered artificial intelligence company Anthropic to respond to allegations that it submitted a court filing containing a ‘hallucination’ created by AI as part of its defense against copyright claims by a group of music publishers. A lawyer representing Universal Music Group, Concord and ABKCO in a lawsuit over Anthropic’s alleged misuse of their lyrics to train its chatbot Claude told U.S. Magistrate Judge Susan van Keulen at a hearing that an Anthropic data scientist cited a nonexistent academic article to bolster the company’s argument in a dispute over evidence.”

EUIPO releases study on generative artificial intelligence and copyright — The U.S. Copyright Office was not the only public body to release a major report on AI and copyright this week. On Monday, the European Union Intellectual Property Office published a comprehensive study on the topic, the purpose of which “is to deepen the general understanding of GenAI’s technical functioning, as well as existing and developing solutions underlying the application of EU rules on copyright and Artificial Intelligence. The study offers an in-depth analysis of GenAI developments from the perspective of EU copyright law, covering technical, legal, and economic aspects.”

« Older articles