The Big Amici Party in the OpenAI Copyright Infringement Litigation

Dance at Le Moulin de la Galette, Pierre-Auguste Renoir (1876). 

Recently, we filed an amicus brief in In re OpenAI Copyright Infringement Litigation, explaining why “market dilution,” a novel theory of harm proposed by plaintiffs, is contrary to copyright principles and precedents. 

This lawsuit matters to authors because they build on existing copyrighted works, so fair use needs to keep protecting learning, research, and creative reuse. As we wrote in our amicus brief: 

It may be tempting to regard market dilution as an AI-specific response to an AI-specific problem of scale. But a legal rule does not stay attached to the technology that prompted it.

The lawsuit has also drawn 55 other amici to the party, which likely sets a record for the number of copyright amicus briefs filed at the trial court level. Below, I give a brief introduction to the plaintiffs and put some amicus briefs in a conversational thread on some key issues that matter for authors.

Who is bringing the lawsuits? 

The OpenAI Copyright Infringement Litigation consolidates copyright cases brought by news media, publishers, and individual authors across multiple federal districts. Together, these rightsholders are attempting to put OpenAI and Microsoft on the hook for potentially billions of dollars, plus statutory damages. The news plaintiffs include the New York Times, represented in part by Susman Godfrey, and The Center for Investigative Reporting and The Intercept, represented by Loevy & Loevy. 

Authors and other rightsholders make up the proposed “book class,” which hasn’t been certified yet. Some authors in the class include George R.R. Martin (author of A Game of Thrones) and Jodi Picoult (author of My Sister’s Keeper). The Authors Guild, a longtime advocate for strong copyright protection, is also a named plaintiff. The proposed book class is led by Susman Godfrey (again), and Joseph Saveri Law Firm, which filed some of the earlier author suits, is also counsel for the book class. ​​Many of the plaintiffs’ lawyers here are among the most active firms bringing AI copyright suits in the country. 

Ziff Davis, a digital media publisher, is suing on its own rather than as part of a class. Aside from bringing traditional copyright infringement claims, Ziff Davis also unsuccessfully tested § 1201 under the Digital Millennium Copyright Act (banning unauthorized circumvention of technological measures) against OpenAI. The court dismissed that claim, finding that “robots.txt directives are merely requests and do not effectively control access to copyrighted works.” 

Putting amicus briefs in conversation

The flood of amicus briefs — totaling 56 — comprises 29 briefs for plaintiffs, 19 for defendants, and 8 in support of neither party. Chamber of Progress helpfully sorts the briefs by issue — transformativeness, market harm, good faith, piracy, and licensing market — and lays out each side’s key arguments. 

Re:Create takes an “artisanal” approach, grouping the amici by who they are and what they want, with labels like “Fair Use Stalwarts” and “Protectionist Stalwarts.”

Since selection and arrangement is often only thinly protected under copyright, I feel pretty comfortable borrowing some of Re:Create’s groupings and labels of amicus briefs. In this post I also want to make some of them talk to each other, so readers have a sense of not only where the amici sit but also how their arguments interact and clash with each other. 

Is “market dilution” a cognizable injury? 

The theory of market dilution, first articulated by a court in Kadrey v. Meta, posits that the fourth factor — assessing market effect on the original work because of the secondary use — should weigh in favor of rightsholders because AI-generated work will crowd out human-authored content, and accordingly, “dilute” the markets for human-authored works. According to Chamber of Progress’s issue-by-issue breakdown, “market dilution” drew comment from at least 44 of the 56 amici, second only to “licensing market”. 

On the protectionist stalwarts side, Association of American Publishers warned that “publishers risk losing revenues and market visibility to the flood of AI-generated books and other texts built on the backs of their own works.” And labor unions in the creative and entertainment industries quoted from an early OpenAI employee’s prediction: “[t]he better we do on GPT-X, the more worried genre fiction authors will become about us substituting for them on Amazon.” 

The fair use stalwarts (a group we are happy to be included with) answered that this worry, however real, isn’t one copyright law addresses. Authors Alliance called market dilution “ordinary competition,” arguing that factor four “reaches uses that substitute for protected expression, not new works that compete for the same readers.” Electronic Frontier Foundation (EFF) made a similar point regarding the scope of copyright: if readers buy non-infringing AI-assisted romance novels instead of Bridgerton, “Julia Quinn has no right under copyright law to prevent these competing books from entering the market.” 

Open Markets Institute, a pro-competition think tank, weighed in for the news plaintiffs, arguing that OpenAI’s unlicensed copying is itself a competition problem. “If an LLM developer copies the Publishers’ content without first securing a license,” it wrote, “they are engaging in unfair and socially destructive competition.” In its view, the harm runs two ways. Readers who get the news free from a chatbot “have no incentive to click through” to news websites. And AI rivals that pay for licenses are left choosing “either accepting a higher cost structure relative to their scofflaw rivals or emulating their lawbreaking.” 

Whether AI developers who license their training data are actually put at a competitive disadvantage is an empirical question, but the brief supported that claim with a prediction instead (licensed developers “may steadily lose market share”) rather than evidence. In any case, it is too early to call unlicensed training “lawbreaking”: whether it carries copyright liability at all depends on how courts resolve fair use. 

The first point, that readers getting free news from AI would lack incentive to visit the publishers’ page, is also an empirical statement. It is an application of the market dilution theory to the news industry that, according to copyright professors Pamela Samuelson and Christopher Sprigman, calls for a showing of empirical evidence. They maintained that even if market dilution is conceptually sound (which they doubted, for reasons similar to those in Authors Alliance’s brief), courts “should not infer that dilution has occurred or is likely.” Instead, they suggested that “[p]laintiffs must identify the markets they claim are harmed and prove that harm with empirical evidence.”

The framing of market dilution as a more general matter is on weak footing too. As Samuelson and Sprigman pointed out, the theory treats the copyright markets as “binary” in the sense that “wholly ‘AI-authored’ works displace wholly ‘human-authored’ works.” But on the horizon is the emergence of rapidly growing “hybrid” works — works created by humans using AI. And a sensible version of market dilution theory, one that the plaintiffs’ version fell short of, should account for hybrid works. 

What is the right forum to address “market dilution”? 

Despite all of its conceptual and empirical shortcomings, market dilution caused by AI, if real, poses unprecedented challenges to authors and creators of almost all kinds. To the extent that market dilution concerns displacement of human authors by AI, “no plaintiff or group of plaintiffs can adequately represent the interests of many creators who are not plaintiffs in those cases but would nonetheless be affected by the outcome of the plaintiffs’ claims,” as Samuelson and Sprigman wrote. On the defendants’ side, they explained that “no single generative AI developer defendant could plausibly be responsible for dilution on its own, because all developers of generative AI systems would likely have contributed to this sort of market-wide harm.” Congress, therefore, not courts, is best suited to address such harm at scale. 

Authors Alliance agrees with Samuelson and Sprigman’s proposal to leave market-wide dilution harm to Congress. How to protect human authors from potential generative AI harms at scale is a policy choice for Congress, which, unlike courts bound to the parties before them, can weigh the interests of everyone affected and enact forward-looking legislation to protect the interests of creators and authors. 

Different angles under the First Amendment 

Copyright fuels creative expression, and in doing so builds a cultural commons available for all to use, in service of the First Amendment’s guarantee of free expression. In the Supreme Court’s words in Harper & Row v. Nation Enterprises, copyright is the “engine of free expression.” 

Copyright and First Amendment scholars Edward Lee and Rebecca Tushnet argued that market dilution theory violates the First Amendment because it penalizes non-infringing works based on their content (AI-generated works) and class of speakers (people who use AI to create). It will be interesting to watch how, if at all, the court takes up this First Amendment argument. 

Another cohort of First Amendment scholars, including Jack Balkin and Burt Neuborne, filed in support of the news plaintiffs, arguing in the other direction and observing that because journalism is important to the democratic process, AI companies’ unlicensed use of news content threatens democracy by undermining the press’s economic viability. 

In particular, America’s Newspapers, a trade association of more than 1,500 local news publishers, warned that the threat is most acute for local news: “National publishers are poised to weather this substitution. Local publishers will not survive it.” The local industry is already shrinking. In the past year, 170 local newspapers have closed or merged. As local outlets keep contracting and AI pulls readers away, the news market risks consolidating around a handful of national publishers. 

Authors Alliance fully recognizes the threat facing local news and what its loss would mean for accountability and democratic discourse. But as with market dilution, we think Congress, not the courts, is the right institution to address industry-wide problem of this scale. And copyright law is likely not the right mechanism to make such adjustments. 

How courts deal with the First Amendment issues matters to authors because they both produce and benefit from the use of news materials. An imbalanced protection of the news industry could shrink the cultural commons and curb authors’ ability to draw from it and create new works. But an over-protective stance would deprive us all of access to the underlying facts and ideas we need to speak freely. 

Does bad faith and unfair dealing taint fair use? 

The idea that bad faith and unfair dealing would taint the first fair use factor — the purpose and character of the use — was first embraced by the Supreme Court in Justice O’Connor’s majority opinion in Harper & Row. She wrote: “Also relevant to the character of the use is the propriety of the defendant’s conduct. Fair use presupposes good faith and fair dealing.” (internal quotation marks and citations omitted).

Since then, the Supreme Court has backed away from understanding “fair” in fair use as “fair dealing”. In Campbell v. Acuff-Rose, the Court said, “If the use is otherwise fair, then no permission need be sought or granted.” In Google v. Oracle, the Court again voiced skepticism, quoting Judge Leval that “[c]opyright is not a privilege reserved for the well-behaved.” The Court declined to decide whether good faith matters at all, calling it a “factbound consideration” that was “not determinative” in that case. Since Campbell and Warhol, whether a secondary use is transformative dominates the analysis of the first fair use factor. 

Recently, plaintiffs in AI copyright litigation sought to revive the bad faith fair use argument, and it had mixed results. In Bartz v. Anthropic, though the court didn’t decide the case on this basis, it suggested that “piracy of otherwise available copies is inherently, irredeemably infringing even if the pirated copies are immediately used for the transformative use. . .” Kadrey v. Meta went the other way. There, the court indicated that “downloading [from a shadow library] must still be considered in light of its ultimate, highly transformative purpose: training Llama.” And because Meta’s end use for the purposes of training its AI model is transformative, “so too was Meta’s downloading of those books.”

Plaintiffs in In re OpenAI Copyright Infringement Litigation tried to marshal similar arguments. The book plaintiffs brought a separate infringement claim for downloading through pirate sites, and the news plaintiffs contended that defendants’ disregarding terms of use, among other things, should weigh against fair use.  EFF urged that fair use should focus on the ultimate use, that the concept of good faith and fair dealing has no place in modern fair use jurisprudence, and that Congress would have made “lawful acquisition” a requirement for fair use if it had so intended. 

Similar to EFF’s last point regarding congressional intent is the amicus brief filed by two former members of Congress, Rick Boucher, a longtime fair use advocate while in Congress, and Rick Santorum, who, as the brief described, understand “both the value of independent journalism and the promise that new tools hold for the press and its audiences.” They put in a footnote that using improper means to obtain copyrighted materials “weighs in favor of the copyright holder but should not necessarily defeat a finding of fair use in all circumstances.” 

Whether and how much fair use accounts for good or bad faith is important to authors because sometimes the most valuable material for investigative journalism and critical biography is material its owner never wants to share. For example, in NXIVM Corp. v. Ross Institute, the Second Circuit held that breaching a nondisclosure agreement to obtain critics’ quotations from NXIVM’s confidential seminar manual — a conduct in “bad faith”— was likely fair use anyway “in light of the transformative nature of the secondary use as criticism.” 

Licensing market 

Chamber of Progress’s issue-by-issue breakdown of amicus briefs shows that the issue of “licensing market” is the most argued one, with at least 47 amici weighing in out of 56 as a whole. Several doctrinal, practical, and policy questions are raised under this banner.

Can licensing work at internet scale? 

Human Consent Foundation and Cate Blanchett, with support from Hollywood stars such as Anne Hathaway and Scarlett Johansson, suggested: “The creative industries already have in place a proven infrastructure and workflows to clear rights for dispersed works at scale and cost-effectively.” Moreover, AI Coalition for Data Integrity sought to show that the licensing market already exists because it already “participate[s] in the licensing market that Defendants say cannot work at scale.” Whether an existing licensing market exists for AI training materials is a question of fact.  

Matthew Sag, a legal scholar in AI and copyright, argued that such a licensing market at scale is impossible because “although training data is valuable in aggregate, the marginal contribution of any individual piece of training data, whether a book or a social media post, is approximately zero.” Midjourney and Runway, two generative AI companies working with images and videos, made a related point about transaction costs: even if each license could be negotiated, “in the span of a single minute—which is, of course, absurd—licensing a million works would take a million minutes, or roughly 1.9 years, and licensing a billion would take nearly two thousand years.” 

Is counting lost licensing fees circular? 

One copyright principle is that mere loss of licensing opportunity cannot be counted as market harm because otherwise the fourth fair use factor (market harm imposed by the secondary use on the original work) would always favor the rightsholder. This is called the circularity problem (see, e.g., American Geophysical Union v. Texaco and Ringgold v. Black Entertainment Television). Instead, the licensing market alleged must be either existing or likely to be developed. 

On this issue, amici on both sides didn’t seem to dispute how courts should interpret the law. A.I. Progress and the Computer and Communications Industry Association urged the court to look only at the sort of market recognized under the fourth factor. Recording Industry Association of America and Copyright Alliance countered with a factual disagreement, arguing that there is no circularity problem because a licensing market already exists. 

Who benefits from a licensing requirement? 

Imposing a license requirement on AI model developers implicates competition issues. Both EFF and Public Knowledge warned that doing so would favor the largest technology companies, since they are most likely to afford licensing fees for millions of copyrighted works. The upshot is that it would be cost-prohibitive for emerging AI companies to level up and effectively compete with the big players. 

Because of the high transaction costs of working with many smaller rightsholder organizations, licensing may also lead to a competitive imbalance among that group as well. For example, a coalition representing the news outlets, including Rebuild Local News, contended that now only large outlets get licensing deals, “resulting in our smallest local news proprietors simply donating their time and labor to Silicon Valley.” 

Creative and expressive activities are not the only kind of human labor implicated by a chatbot with increasing capabilities. Software engineers, Uber drivers, translators, to name a few, are potentially facing market dilution by AI or AI-assisted automation, too. The collective disruption, dilution, and even displacement of human workers by AI across industries is a labor problem, a problem that is neither intrinsic to copyright law nor one it is capable of addressing. Rather than leaving courts to stretch copyright doctrine into a tool it was never meant to be, other legal frameworks and potentially legislation to deal with these risk and related competition issues is the better approach.


Discover more from Authors Alliance

Subscribe to get the latest posts sent to your email.

Leave a Comment

Scroll to Top