Friends of the Court Flock to the OpenAI Copyright Fight

Precedent network

1. Following the citations in the OpenAI copyright amicus briefs

Each line connects an amicus brief in In re: OpenAI, Inc. Copyright Infringement Litigation to a court decision it cites. Briefs supporting plaintiffs sit on the left, briefs supporting OpenAI and Microsoft on the right. Each case is placed by the share of its citing briefs on each side. Hover or tap to trace the connections; click to pin; Shift-click a second brief to compare two briefs.These are the court decisions most often cited by the amicus briefs in In re: OpenAI, Inc. Copyright Infringement Litigation. Each dot sits on a line running from briefs supporting plaintiffs, on the left, to briefs supporting OpenAI and Microsoft, on the right, placed by the share of its citing briefs on each side. Tap a case to see which briefs cite it.

The plaintiffs are two groups, authors in the book cases and news organizations, and some briefs support only one of them; some briefs likewise support OpenAI but not Microsoft. Citing a case is not the same as relying on it; a brief may cite a precedent to argue it does not apply.

Brief supporting plaintiffs Brief supporting defendants Neither party No side declared Court decision (size and shade = briefs citing it)

How we counted. Case citations were extracted from each brief's table of authorities and its text with the Free Law Project's eyecite parser, parallel citations were merged, and each brief's citations of these cases were checked against a search for the case's name. Cases were matched to CourtListener's citation database where available. The table view also counts references, every time a brief's text cites a case, including short-form citations and "Id.," as matched by eyecite; a sample of "Id." matches was checked against the briefs. Shown: the cases cited by at least eight of the 56 briefs Judge Sidney H. Stein accepted on Nov. 6, 2026. Source: Tech Policy Press analysis.

How this was made. This analysis was conducted with Anthropic's Claude Opus 5.5, an AI model, with prompts from Tech Policy Press editors, who set the rules for categorizing the briefs. Claude wrote and ran the code that downloaded the briefs and extracted their citations, read each brief, and decided each ambiguous citation individually after reading it in context. Those decisions were not reviewed by an editor; each is logged, with its reason, in the downloadable data. A Tech Policy Press editor checked a sample of the quotes in the downloadable data against the briefs and found each one verbatim.

Amicus tracker

2. What the briefs say, in their own words

All 56 amicus briefs are listed below with the side each one declares; filter by side, and open “In their words” to read verbatim quotes from each brief.

Briefs accepted Nov. 6Word limit 6,000Replies due Nov. 20Oral argument No date set

• ECF 2099

  The AI Rights Project

  Supports plaintiffsFiled Oct. 14Proposed brief, 24 pages

  In their words 6 quotes

  “The AI Rights Project (‘AIRights’) is an independent, nonpartisan 501(c)(3) public-interest organization applying its Human Baseline Principle — a framework for identifying when AI systems displace the human-scale assumptions embedded in existing legal doctrine — to precedent-setting AI litigation.”

  • “Transformative-use doctrine developed without having to state the human creativity predicate, because every actor it evaluated was human.”
  • “The gap between the accurate description and the characterization is the equivalence assumption — and the opinion never examines whether it holds.”
  • “AI training presents the pairing of expression-seeking copying with the absence of human judgment relating any work to a new purpose.”
  • “The unprotectability of style, like the transformativeness of training, was settled in a world where the only learner was human. Neither settles what follows when the learner is not.”
  • “As the proponents of an affirmative defense, Defendants must establish this premise, and this record does not.”
  • “AIRights urges no outcome. We ask only that the threshold question be examined rather than assumed: before machine-scale training is treated as equivalent to the human creative reuse the doctrine was built to evaluate, the Court should examine whether that predicate is satisfied on this record.”
• ECF 2111

  Kevin Kelly

  Supports defendantsFiled Oct. 16Proposed brief, 24 pages

  In their words 6 quotes

  “Mr. Kelly is the founding executive editor of Wired magazine. He is also a New York Times bestselling author, futurist, and photographer who writes fiction and nonfiction about the future of technology.”

  • “First, AI is not trained or designed to create copies of original works. Rather, AI mimics the way humans have learned and innovated for millennia, by ingesting and integrating information into a broader structure of knowledge derived from a vast number of inputs.”
  • “Beyond literature, AI is advancing drug discovery, solving long-unsolved mathematical problems, and cracking cold criminal cases—all without infringing upon any copyrighted interests as traditionally understood.”
  • “Thus, AI, and the training that creates it, is a technological tool that helps authors create new, high-quality original works.”
  • “In other words, Plaintiffs seek to prevent other authors from using AI as a creative tool to write non-infringing works that compete with Plaintiffs’ works.”
  • “Amicus recognizes serious questions about AI’s technical capabilities and its effects on society. Copyright law does not answer all of them. It asks only whether allowing the use better serves the progress of science and the useful arts than preventing it would. AI training does.”
  • “This Court should hold that Defendants’ use of copyrighted works to train AI models is a fair use under 17 U.S.C. § 107.”
• ECF 2283

  Ron Suskind and Brian L. Frye

  Supports neither partyFiled Oct. 28Proposed brief, 13 pages

  In their words 6 quotes

  “Ron Suskind is a Pulitzer Prize-winning journalist, author, and filmmaker. Suskind is also the CEO of Sourcebase, an AI company founded and led by journalists.”

  • “This amicus brief observes that generative AI does not necessarily harm the market for or value of news, and may in fact benefit the News Plaintiffs by reducing their costs and increasing the value of their brands.”
  • “But none of those harms are relevant, because they were all the result of fair competition, which copyright is supposed to encourage, not prohibit.”
  • “On the contrary, using newspaper articles to train a generative AI model should be a fair use, so long as the model attributes its output to the newspaper articles on which it was trained.”
  • “There is no authority to support a dilution theory of copyright harm, because dilution is not a copyright harm.”
  • “If AI companies are forced to license newspaper articles in order to train their generative AI models, only the largest and wealthiest AI companies will be able to compete.”
  • “As the Supreme Court observed in Sony, the courts should leave it to Congress to decide whether and how to regulate this revolutionary new technology.”
• ECF 2304

  Association of American Publishers, Copyright Clearance Center, Independent Book Publishers Association and International Association of Scientific, Technical & Medical Publishers

  Supports plaintiffsFiled Oct. 29Proposed brief, 25 pages

  In their words 6 quotes

  “The Association of American Publishers, Inc. (‘AAP’) is a not-for-profit organization that represents approximately 120 book, journal, and education publishers in the United States on matters of law and policy ...”

  • “LLMs like OpenAI’s exploit copyrighted works for their expressive value for the same purpose, ultimately generating substitutes for the works used for training.”
  • “There is no apparent, let alone compelling, justification for AI training on expressive content to generate expressive content.”
  • “Consumers are confused and overwhelmed, and 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.”
  • “Here, a market for licensing copyrighted works for AI exists and is expanding to address growing demands and developing infrastructure.”
  • “AI companies’ use of books and other copyrighted text from illegal sources normalizes large-scale theft by supporting and promoting the expansion of piracy as a business.”
  • “The Court should grant the plaintiffs’, and deny the defendants’, motions for summary judgment and find that OpenAI’s unauthorized use of copyrighted works for AI training is not fair use.”
• ECF 2306

  Center for American Rights, Article III Project, Internet Accountability Project and others

  Supports plaintiffsFiled Oct. 29Proposed brief, 18 pages

  In their words 6 quotes

  “The Center for American Rights is a nonprofit, nonpartisan public interest law firm with a focus on issues related to broadcasting, communications, telecom, and technology.”

  • “This, of course, is the Plaintiffs’ collective point: the SI companies can access the information they seek, they just need to pay fair and appropriate remuneration.”
  • “Defendant’s view complicates the Trump Administration’s vision in one critical way: it seeks an unqualified ability to subvert the rights of content owners.”
  • “To ensure diversity in our news and information ecosystem, this court must ensure that intellectual property is maintained and not overwhelmed by SI.”
  • “However, IP theft is not only an issue of monetary loss—our gravest foreign adversaries also distill American models to steal capabilities that would otherwise constitute copyright violations.”
  • “We are okay with seeing mainstream news media organizations shrink, but because they deserve declining readership for their fake news, not because their IP has been stolen.”
  • “If the Court here agrees with OpenAI’s expansive view of fair use, it would not only encourage foreign adversaries to continue this obvious form of IP theft but also give them legal immunity to do so. As a matter of equity, these national security concerns counsel against adopting such an interpretation.”
• ECF 2308

  Matthew Sag

  Supports neither partyFiled Oct. 29Proposed brief, 26 pages

  In their words 6 quotes

  “I am the Jonas Robitscher Professor of Law in Artificial Intelligence, Machine Learning, and Data Science at Emory University School of Law.”

  • “On the record in this case, the defendants’ use of plaintiffs’ works for AI training is highly transformative and strongly favors a finding of fair use because under realistic conditions the outputs of defendants' models are in no relevant way similar to the invisible intermediate copies made during the training process.”
  • “Defendants’ models do not, in their ordinary operation, deliver the plaintiffs’ expression to anyone. The parties dispute how regurgitation in ChatGPT conversations should be counted, but on any measure it is rare.”
  • “The problem in a nutshell is that 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.”
  • “Copyright is a system of individual rights, but the harms asserted under the banner of dilution are collective harms.”
  • “The relevant question is whether, by obtaining books as it did, OpenAI bypassed a market for access that was actually available to it in a way that was likely to cause harm.”
  • “Accordingly, amicus respectfully submits that the Court should hold that training the models at issue on the plaintiffs’ works was transformative; and determine whether OpenAI’s acquisition of books from shadow libraries is an issue that survives summary judgment on this record.”
• ECF 2313

  Public Knowledge

  Supports defendantsFiled Oct. 29Proposed brief, 26 pages

  In their words 6 quotes

  “Public Knowledge is a nonprofit public-interest organization dedicated to preserving an open internet and the public’s access to knowledge, promoting creativity through balanced intellectual property policies, and protecting consumers’ ability to use innovative technology lawfully.”

  • “Public Knowledge therefore urges the Court to consider the pivotal issue of AI training separately from data acquisition and AI’s outputs.”
  • “Training is a highly transformative activity: it analyzes works to calculate statistical relationships among tokens. It does not provide copies of those works, or of protectable expression, to readers as a substitute for the original works in the marketplace.”
  • “While copyright protects the works’ expression, it does not give authors a monopoly over learning or using objective information about their works. Learning these patterns is distinct from reproducing a work’s protected sequence of words.”
  • “The Court should therefore reject a ‘market dilution’ theory based on AI models (after training) facilitating the creation of new, noninfringing works. An author’s copyright does not reserve the market for similar works created by others.”
  • “A finding that model training is not fair use, by contrast, could favor the largest technology and media companies.”
  • “For the foregoing reasons, the Court should hold that the challenged training use is a fair use.”
• ECF 2325

  Digital Content Next

  Supports plaintiffsFiled Oct. 29Proposed brief, 23 pages

  In their words 6 quotes

  “Founded in 2001, DCN is the only trade organization dedicated to serving the unique and diverse needs of high-quality digital content companies that enjoy trusted, direct relationships with consumers and advertisers.”

  • “AI tools that copy content and present it to users without sending them directly to the source (or without other financial remuneration) strike at the heart of publishers’ business models and deprive them of revenue necessary to produce new works.”
  • “Thus, RAG makes it unnecessary for a user to ‘click through’ to read and engage with the external source material itself.”
  • “It found that, when compared to the YoY web traffic in 2025, the YoY web traffic in 2026 had declined even more dramatically, with a median decline of 22 percent.”
  • “This Court should accordingly conclude that Defendants’ use of Plaintiffs’ works fails to constitute ‘fair use.’”
  • “This Court should not interfere with this burgeoning market by implying a royalty-free license that enriches AI companies at the expense of digital publishers.”
  • “Defendants’ misuse of publishers’ data is not ‘fair’, and it is not legal—and this Court should not countenance it.”
• ECF 2328

  Authors Alliance

  Supports defendantsFiled Oct. 29Proposed brief, 25 pages

  In their words 5 quotes

  “Authors Alliance is a nonprofit organization that advances the interests of authors who write to be read and serve the public good.”

  • “A theory that counts the creation of new, noninfringing works as a harm to be weighed against fair use would discourage the very abundance of expression copyright exists to promote.”
  • “Copyright law does not police competition; it polices substitution for protected expression.”
  • “Scale can magnify a harm that copyright recognizes, but it cannot create one where none exists.”
  • “Treating lawful new speech as an injury because it competes with earlier copyrighted speech would press copyright’s exclusive rights into territory those built-in safeguards are designed to keep open.”
  • “A rule broad enough to protect an incumbent author from noninfringing AI-generated competition is broad enough in principle to protect an incumbent author from noninfringing human competition that fair use helped enable.”
• ECF 2332

  SAG-AFTRA, American Federation of Musicians, Writers Guild of America West and Writers Guild of America East

  Supports plaintiffsFiled Oct. 29Proposed brief, 34 pages

  In their words 6 quotes

  “Amici Curiae (the ‘Unions’) are labor unions representing working professionals in the news and entertainment industries.”

  • “Here, the purpose of using human-created works is to train a model to generate works that emulate and compete with human-created works.”
  • “Memorization is not an isolated defect in an otherwise transformative use; memorizing human expression is at the core of how Defendants’ models are trained.”
  • “These negotiated provisions presuppose that training on covered work is a use that can be licensed and that the people whose work is used have a stake in the terms. A ruling here that training is categorically fair use would render moot the terms the Unions and the industry agreed to at the bargaining table.”
  • “These AI licensing markets are young and imperfect. But the answer to a young market is to let it develop, not to declare the use free.”
  • “The outcome here will reach well beyond Plaintiffs’ works, including to those on which the Unions’ members build their livelihoods. The Court should therefore reject any categorical rule.”
  • “For the foregoing reasons the Court should reject Defendants’ fair use defenses.”
• ECF 2335

  Engine Advocacy, Developers Alliance and Lazy Moose Co.

  Supports defendantsFiled Oct. 29Proposed brief, 25 pages

  In their words 6 quotes

  “Engine Advocacy (‘Engine’) is a nonprofit technology policy, research, and advocacy organization dedicated to bridging the gap between startups and policymakers.”

  • “Training is transformative, and it takes nothing from the market for the works on which models learn.”
  • “Startups are not using AI foundation models to create copies of the original, copyrighted works to compete with those works in the market. They are creating tools that generate completely new outputs that in no way substitute for the original works.”
  • “Licensing the works needed to train a single large-language model would cost on the order of $10 trillion, and developers passing those costs downstream in whole would need to charge each commercial user roughly $1.4 million—about the size of an average startup’s entire first funding round, and an impossible sum for a seed-stage company.”
  • “If using copyrighted material for training data is infringement and not a fair use, it would effectively put an end to open-source foundation models, which startups rely on heavily.”
  • “Billions of potentially copyrightable works are needed to train a single large-language model. The rightsholders for all those works cannot realistically be identified, let alone negotiated with one by one.”
  • “For the foregoing reasons, the Court should hold that the use of copyrighted works to train AI models is a fair use and grant Defendants’ motions for summary judgment on that issue.”
• ECF 2359

  Copyright Alliance

  Supports plaintiffsFiled Oct. 29Proposed brief, 24 pages

  In their words 6 quotes

  “The Copyright Alliance submits this amicus brief because our members have a strong interest in proper application of the Copyright Act, especially the application of copyright’s fair use doctrine to generative artificial intelligence (GAI).”

  • “Warhol, Texaco, and many of the cases discussed below make clear that, when considering the purpose of the use in a factor-one fair use analysis, the analysis must not merely consider the intermediate step but rather must take into account the ultimate purpose of the use.”
  • “There is nothing speculative or theoretical about the robust and ever-expanding licensing market for AI training.”
  • “But Warhol dramatically changed the landscape, both narrowing what ‘transformative’ means and clarifying that a finding of transformativeness does not lead ineluctably to a conclusion that the use was fair.”
  • “The reality is that GAI models ingest and reproduce perfect copies of copyrighted works (and generate material) at a speed and scale that makes any comparison to human learning utterly inapt, and Defendants’ attempt to liken the two should be roundly rejected.”
  • “However, downloading illegal copies of works from known piracy repositories is not immunized by a use that may come after.”
  • “This Court should deny Defendants’ motions for summary judgment.”
• ECF 2361

  Wikimedia Foundation

  Supports neither partyFiled Oct. 29Proposed brief, 27 pages

  In their words 6 quotes

  “Wikimedia Foundation, Inc. (‘Wikimedia’) is a non-profit charitable foundation based in San Francisco, California ...”

  • “Scraping is simply a means of acquiring copies of works, but acquisition in and of itself is not a use of those works.”
  • “... the experience that Wikimedia and its volunteer communities have with building and using AI/ML tools suggests that using copied works to build a general purpose LLM is likely transformative and does not inherently create market harm by substituting for the underlying works.”
  • “It is likely not fair use when an LLM-based service uses copies of original works for their intended purposes—to provide relevant information, perspectives, expressions, and context—while supplanting readers’ engagement with those original works.”
  • “Users diverted from visiting and thus benefiting from the information available on a Wikipedia entry are also diverted from the opportunity to further that knowledge via, for example, volunteer editing or donating.”
  • “Compliance with the provisions of Creative Commons or other copyright licenses would make RAG uses of works noninfringing.”
  • “Specifically, Wikimedia respectfully encourages the Court to adopt the following positions: 1) what is done with scraped copies – and not the act of scraping – should be treated as the ‘use’ that is analyzed; 2) the pre-training of ChatGPT’s LLM models is fair use; 3) the copying and reproduction of expressive sources in responses to user queries by ChatGPT’s RAG system is not fair use.”
• ECF 2369

  The Copia Institute (Floor64)

  Supports defendantsFiled Oct. 30Proposed brief, 24 pages

  In their words 6 quotes

  “Amicus Copia Institute is the think tank arm of Floor64, Inc., the privately-held small business behind Techdirt.com (‘Techdirt’), an online publication that has chronicled technology law and policy for nearly 30 years.”

  • “Copyright law does not and cannot do what the plaintiffs seek it to do: obstruct the development of AI by targeting its training.”
  • “Thus, given that the sole purpose of copyright law is to promote innovation, it cannot be a law that is used to obstruct it.”
  • “Even if copying may technically occur during training, it is incidental to what the defendant is really doing: learning, just as anyone learns from consuming a work.”
  • “Read, watched, listened to, learned from. These are the verbs implicated by AI training, and these are the activities that the First Amendment, via fair use, protects—and copyright itself exists to ensure can happen.”
  • “Thus it is critically important that copyright law not interfere with AI training in order not to interfere with the public’s ability to consume works as the First Amendment says they have the right to do.”
  • “The defendant’s development of AI does not violate copyright law.”
• ECF 2370

  A.I. Progress

  Supports defendantsFiled Oct. 30Proposed brief, 31 pages

  In their words 6 quotes

  “Amicus curiae A.I. Progress, Inc. is a coalition of leading AI developers committed to the advancement of AI tools that can deliver benefits across societies and economic sectors, from public safety to scientific discovery.”

  • “The sheer range of these beneficial uses confirms that the purpose of AI training—a key consideration in the fair-use inquiry—is not to generate substitutes for authors’ works.”
  • “Generative AI training is a transformative use because it does not use copyrighted works to communicate their expression.”
  • “The possibility that a downstream user might prompt a model to produce an infringing output cannot determine whether the antecedent copying undertaken to train the model is itself fair use. The Court must evaluate training for what training does and the purpose it serves.”
  • “But evidence of commercial demand for licensed display and product uses does not establish a cognizable licensing market for the distinct and transformative use of copyrighted works in AI training.”
  • “In other words, the theory does not identify harm resulting from the exploitation of copyrighted expression; it identifies harm resulting from increased competition. Copyright law has never recognized the latter as a cognizable injury.”
  • “The Court should reject Plaintiffs’ attempt to extend copyright beyond its established limits. AI training uses copyrighted works for a fundamentally different purpose, does not provide the public with substitute access to those works, and does not cause the market harm cognizable under the Copyright Act.”
• ECF 2373

  News/Media Alliance, National Newspaper Association, National Press Photographers Association and ProPublica

  Supports plaintiffsFiled Oct. 30Proposed brief, 26 pages

  In their words 6 quotes

  “News/Media Alliance (‘NMA’) is a nonprofit organization representing over 2,200 publishers in the United States, ranging from the largest news and magazine publishers in the country to hyperlocal newspapers, and from digital-only outlets to papers that have printed news since the nation’s founding.”

  • “Defendants’ copying of news content to train and then operate their AI systems using RAG is not transformative under the first fair use factor because the use has the same purpose as, and serves as a substitute for, original journalistic works.”
  • “RAG is thus a factually distinct form of unauthorized copying by Defendants beyond the original training process that fails the fair use test in its own right.”
  • “AI news chatbots and the like allow users to consume publishers’ work product without the need to visit the sites from which it was taken.”
  • “A determination that news content can be taken with impunity would threaten news publishers’ economic viability by eviscerating a well-established licensing market for news content, ultimately diminishing the availability of high-quality journalism to the public.”
  • “A ruling in this case that makes journalistic works free for the taking so trillion-dollar companies can profit by offering competing substitutes will put news publishers out of business and cannot be reconciled with the constitutional vision or public interest.”
  • “For the foregoing reasons, Amici respectfully urge this Court to grant Plaintiffs’ motion for partial summary judgment and to deny Defendants’ motions for summary judgment.”
• ECF 2377

  Open Markets Institute

  Supports plaintiffsFiled Oct. 30Proposed brief, 25 pages

  In their words 6 quotes

  “The Open Markets Institute (‘OMI’) is a non-profit organization dedicated to protecting democracy and individual liberties from concentrated economic power and control.”

  • “If an LLM developer copies the Publishers’ content without first securing a license, they are engaging in unfair and socially destructive competition.”
  • “When the original and the copy operate in the same market and are used by the audience for the same purpose, they are substitutes, not complements.”
  • “But the Publishers’ economic viability depends on users visiting their websites because they earn revenue through subscription fees and advertising.”
  • “Unless this conduct is enjoined by the court, rival LLM developers that respect copyright protection and seek licenses will be harmed. These firms will face the choice of either accepting a higher cost structure relative to their scofflaw rivals or emulating their lawbreaking.”
  • “In other words, the success of the lawbreakers unleashes a race to the bottom as competitors realize they must also break the law to succeed.”
  • “For the reasons stated above, the Publishers’ combined summary judgment motion should be granted and the Defendants’ motion for summary judgement should be denied.”
• ECF 2378

  Foundation for American Innovation, David R. Shedd, retired Lt. Gen. Michael Groen and George Barnes

  Supports defendantsFiled Oct. 30Proposed brief, 25 pages

  In their words 6 quotes

  “The Foundation for American Innovation is a nonprofit public-policy organization.”

  • “But factor four of the fair use test is not limited to an accounting of the copyright holder’s potential lost revenue.”
  • “Requiring permission for the challenged training could impair those benefits if it materially restricts access to useful training data.”
  • “No single work need independently create translation, coding, synthesis, or reasoning capability for corpus-scale training to produce those capabilities collectively.”
  • “These measures underscore that access to useful data is itself a dimension of AI competition.”
  • “The causal question is whether the challenged training produced capabilities the Government uses. If OpenAI’s record establishes that link, the Court should weigh the resulting national-security benefits alongside cognizable market harm under factor four.”
  • “Amici respectfully request that, under Oracle, the Court weigh under factor four the national-security benefits that the record links to the challenged training, alongside cognizable market harm.”
• ECF 2380

  Pamela Samuelson and Christopher Jon Sprigman

  Supports neither partyFiled Oct. 30Proposed brief, 26 pages

  In their words 6 quotes

  “Pamela Samuelson is the Richard M. Sherman Distinguished Professor of Law at Berkeley Law School and Faculty Co-Director of the Berkeley Center for Law & Technology.”

  • “Market dilution—the prediction that generative AI outputs will flood creative markets and crowd out human-authored works—is a novel and speculative theory of harm.”
  • “Whatever competitive pressure AI-assisted works exert comes from their existence as new works in the market, not from any appropriation of a plaintiff’s expression.”
  • “It makes little sense to understand competition from hybrid works—even hybrid works that involve a great deal of human creativity in their production—as categorically ‘diluting’ the market for works made without AI.”
  • “Early empirical studies testing this novel theory provide some insights into how markets are being affected by AI systems, but the results thus far are equivocal.”
  • “Collective harms require collective solutions. It is Congress, not courts in individual cases, that can best address collective harms.”
  • “For these reasons, amici respectfully urge the Court to reject Plaintiffs’ market dilution theory as inconsistent with copyright law. At a minimum, the Court should decline to infer dilution and should require Plaintiffs to prove it with empirical evidence.”
• ECF 2381

  Joshua S. Gans

  Argument leans toward the defendants.

  Declares no sideFiled Oct. 30Proposed brief, 26 pages

  In their words 6 quotes

  “Joshua S. Gans is a Professor of Strategic Management and holds the Jeffrey S. Skoll Chair in Technical Innovation and Entrepreneurship at the University of Toronto’s Rotman School of Management.”

  • “The News Plaintiffs’ first asserted harm—a reduction in the plaintiffs’ sales—does not by itself demonstrate cognizable harm.”
  • “The fact that a model was trained on a work therefore does not show that the copyrighted work made a given output possible or produced a given harm to the copyright owner.”
  • “But a new use can become valuable before any arrangements for licensing that use exist, so a licensing market visible today does not show that a workable one existed when the training occurred.”
  • “Such an approach will support fair use protection for general-purpose learning where the statutory factors warrant it, while triggering greater scrutiny when conduct is shown to produce substitutes for protected expression or harm to cognizable licensing markets.”
  • “Those features raise their own fair-use questions, and the Court should evaluate them on their own terms.”
  • “The Court should therefore reject the News Plaintiffs’ attempt to categorically preclude a finding of fair use for general-purpose training based on those broad assertions and instead assess supported harm and cognizable markets based on the statutory factors.”
• ECF 2382

  Minderoo Centre for Technology and Democracy, University of Cambridge

  Supports plaintiffsFiled Oct. 30Proposed brief, 24 pages

  In their words 6 quotes

  “Amicus examines how claims about what artificial intelligence systems contain and do are made, tested, and audited, and how those systems affect creative work.”

  • “But while there is no dispute that model ‘training’ is how all parties define what Defendants did with the Asserted Works, there is a substantial and unresolved dispute about what ‘training’ includes, and what happens to the Asserted Works during training.”
  • “A user may well make a transformative use of ChatGPT, but that says nothing about whether Defendants’ own copying and training were fair.”
  • “A copy whose contents travel into the finished product is not intermediate.”
  • “A model can be trained not to emit memorized data only if there is memorized data to withhold, and a low output rate from a product designed to suppress output measures how well the suppression works, not what the model holds.”
  • “That is a genuine dispute of material fact, and it cannot be resolved on summary judgment by crediting one side.”
  • “The Court should deny Defendants’ motions to the extent they seek a ruling that training on Plaintiffs’ works is fair use as a matter of law.”
• ECF 2384

  Consumer Technology Association, American Council of the Blind, American Library Association, Association of Research Libraries and Business Software Alliance

  Supports defendantsFiled Oct. 30Proposed brief, 24 pages

  In their words 6 quotes

  “The Consumer Technology Association (CTA®, www.cta.tech) is North America’s largest technology trade association.”

  • “A rule requiring permission from each copyright owner before training would favor institutions able to assemble extensive rights portfolios and conduct large-scale negotiations.”
  • “These data sets themselves are not ‘expressions.’ They represent numerical relationships that enable AI systems to predict the words and ideas to be output in response to user prompts.”
  • “Reliance only on public domain materials is no solution.”
  • “AI systems could not reach the scope and breadth necessary to avoid biased, skewed, and inaccurate responses if any copyright owner could thwart fair use simply by refusing to license at a fair price, or at all.”
  • “By allowing AI systems to train on vast sets of material, including copyrighted works, fair use makes possible these paradigm shifts in how society works, learns, and creates.”
  • “Wherefore, amici respectfully request that the Court grant summary judgment to Defendants, and deny summary judgment to Plaintiffs.”
• ECF 2385

  AI Coalition for Data Integrity and others, including the Writers Guilds of America West and East

  Supports plaintiffsFiled Oct. 30Proposed brief, 24 pages

  In their words 6 quotes

  “The AI Coalition for Data Integrity (‘AICDI’) is a coalition of technology and content interests that supports transparency about artificial intelligence (‘AI’) training data and the licensing of copyrighted works used in training.”

  • “Properly considered, the security interests of the United States favor a training-data ecosystem that is paid for and documented, not one built on uncompensated and unrecorded copying.”
  • “Every license identifies a work, a source, and a permitted use, so a developer that licenses its data gets a provenance record as a byproduct. A developer that copies without permission has every reason to keep none.”
  • “The nations with whom U.S. companies most directly compete and partner are converging on conditions of lawful access, rightsholder choice, or disclosure. A fair use ruling here would not free U.S. models from those conditions.”
  • “Training data, in other words, has value that Defendants themselves recognize, price, and exchange with one another, while the creators whose works make up those data sets receive nothing.”
  • “And because fair use is an affirmative defense, a developer that cannot document what it copied, from where, and on what terms cannot carry its burden of proving that each use was fair.”
  • “The Court should grant Plaintiffs’ motions for summary judgment and deny Defendants’ motions.”
• ECF 2388

  First Amendment scholars (11, including RonNell Andersen Jones, Jack M. Balkin and Sonja R. West)

  Supports plaintiffsFiled Oct. 30Proposed brief, 31 pages

  In their words 6 quotes

  “Amici are First Amendment scholars and practitioners who have taught courses on constitutional law and the First Amendment, published books and articles on the First Amendment and press-related topics, and dedicated substantial time to the study of First Amendment protections.”

  • “The press performs at least three functions critical to our governmental system: it gathers new facts; it organizes these facts and distributes them to the voting public; and it checks government and corporate power.”
  • “To perform its constitutional role, an independent press requires an economic model that sustains and fuels its independent reporting.”
  • “Defendants built their generative artificial intelligence (‘AI’) products on the constitutionally salient labor of the press.”
  • “Yet Defendants’ chatbot products do not merely exacerbate this peril and accelerate the financial decline of the press; they represent a new type of threat.”
  • “Relying on Defendants’ products to fill the gap left by a disappearing press risks leaving powerful societal actors unmonitored, especially the technology companies responsible for creating these AI chatbots.”
  • “For the foregoing reasons, the Court should grant the News Plaintiffs’ motion for summary judgment and deny the Defendants’ motions for summary judgment.”
• ECF 2393

  Recording Industry Association of America, National Music Publishers' Association and other music groups

  Supports plaintiffsFiled Oct. 30Proposed brief, 25 pages

  In their words 6 quotes

  “Amici curiae (‘Amici’) represent the creators behind much of this country’s most valuable music.”

  • “AI companies copy millions of copyrighted sound recordings and musical compositions without authorization to train their models to generate a limitless supply of synthetic substitutes that invade the music marketplace.”
  • “However, fair use does not exonerate copying that propagates content devoid of human authorship in order to supplant human creation.”
  • “But the relevant purpose is not that of training large language models (‘LLMs’) generally, or of LLMs themselves, but rather of the specific use challenged as infringing – i.e., training on specific copyrighted content for purposes of generating specific kinds of outputs.”
  • “This history proves that licensing on the scale needed for training is operationally and economically feasible.”
  • “Reproduction does not require public exposure of the copyrighted work, or else infringers would always be free to copy in secret. Moreover, AI has the unique capacity to output endless content that displaces the market for and diminishes the value of the copyrighted works on which it trained.”
  • “For these and the other reasons cited above, the Court should find that Defendants’ training on Plaintiffs’ copyrighted works is not fair use as a matter of law.”
• ECF 2394

  Edward Lee and Rebecca Tushnet

  Supports defendantsFiled Oct. 30Proposed brief, 28 pages

  In their words 6 quotes

  “Amici Copyright Law Professors Edward Lee and Rebecca Tushnet are scholars whose research focuses on copyright law and the First Amendment.”

  • “OpenAI’s scaling of datasets with intermediate copies of copyrighted works to develop better, more advanced AI provides ample justification and a highly transformative purpose under Factor 1 of fair use.”
  • “Locating all copyright owners for the millions of training works would be difficult enough; negotiating licenses for them, impossible.”
  • “These noninfringing outputs are not substantially similar to Plaintiffs’ works but are simply in the same genre or contain similar facts. Plaintiffs’ unprecedented theory of ‘dilution’ turns copyright on its head.” (‘dilution’ is the plaintiffs’ term)
  • “Dilution disfavors the noninfringing works of only one class of speakers—people who use Defendants’ AI in creating their works. By discriminating based on both content and class of speakers, dilution violates the First Amendment.”
  • “Plaintiffs may offer evidence showing that they expect to make less revenue than they would in a hypothetical world without AI or with universal licensing, but, under the First Amendment, such a showing is not evidence of a compelling need to insulate them from competition from noninfringing works.”
  • “The Court should grant Defendants’ motion for summary judgment on fair use.”
• ECF 2399

  Ashley Judd

  Supports plaintiffsFiled Oct. 30Proposed brief, 13 pages

  In their words 6 quotes

  “Ashley Judd is an actress, humanitarian, and political activist who was the first prominent person to go on the record as a source in The New York Times’s investigation into Harvey Weinstein.”

  • “Copyright protection makes investigative journalism like the reporting that gave Ms. Judd the confidence to publicly accuse Weinstein possible.”
  • “Allowing artificial intelligence to freeload off of the work of investigative journalists threatens the industry as a whole, and risks leaving potentially paradigm-shifting stories like Ms. Judd’s untold.”
  • “In particular, Defendants’ products threaten the market for the Times’s investigative reporting by offering users a substitute for the reporting itself: not merely the underlying facts, which anyone may report, but the Times’s expression of them.”
  • “The proliferation of generative-AI products will also adversely affect the market for investigative journalism by removing the human element from the storytelling process, making future sources like Ms. Judd more reluctant to go on record for fear that their stories will be retold irresponsibly or inaccurately.”
  • “Generative-AI models trained on a historical record shaped by such disparities risk reproducing them. For a survivor deciding whether to come forward, that risk is one more reason to hesitate.”
  • “The Court should deny Defendants’ motions for summary judgment.”
• ECF 2402

  Authors, photographers, playwrights and literary translators (12 creator groups, including the Dramatists Guild and the National Writers Union)

  Supports plaintiffsFiled Oct. 30Proposed brief, 27 pages

  In their words 6 quotes

  “Amici curiae comprise leading industry associations committed to supporting creators and their rights, including authors, photographers, playwrights, and literary translators.”

  • “To develop these models, Defendants systematically looted millions of works of authorship and exploited their creative expression, without permission from or compensation to creators, knowing that they would destroy creators’ professions.”
  • “Thus, even if the Court finds that Defendants’ use is transformative, such a finding would not outweigh actual market harm or devaluation of amici’s works.”
  • “Like the legal platform in ROSS, Defendants’ models explicitly and by design seek to generate the same types of works as amici, and compete in the same markets that sustain amici’s livelihoods.”
  • “Defendants deliberately exploited the expressive content and value of copyrighted books with the express end goal of competing against, and destroying the market for, those very books.”
  • “The fair use doctrine should not be weaponized to allow Defendants to knowingly exploit ‘purloined’ material with the end goal of driving amici and other similarly situated creators out of business.” (quoting the Supreme Court in Harper & Row v. Nation Enterprises)
  • “For the foregoing reasons, the Court should grant Class Plaintiffs’ Partial Motion for Summary Judgment: (1) that OpenAI’s downloading of pirated copies from LibGen and via Books3 is not fair use; (2) that OpenAI’s further copying of Plaintiffs’ works to train its Large Language Models is not fair use; and (3) that OpenAI and Microsoft’s distribution of copies of Plaintiffs’ works to each other and third parties is not fair use.”
• ECF 2403

  Sarah Rose Siskind and Michael Wall

  Supports defendantsFiled Oct. 30Proposed brief, 21 pages

  In their words 6 quotes

  “Amici curiae are independent creators who use AI tools, including large language models (LLMs), in their creative and commercial work.”

  • “Plaintiffs’ suit oversimplifies the relationship between AI and the creator community, ignoring at least three basic facts about the use of AI among many independent creators.”
  • “AI has helped level the playing field. By allowing independent creators to complete technical and time-consuming business tasks more quickly, AI reduces the competitive disadvantages that independent creators otherwise face.”
  • “AI is not a substitute for creativity. It is a way to augment that quintessentially human undertaking ...”
  • “A rule that lets only the largest companies afford LLM development, and that steers licensing revenue mainly to legacy media outlets, would leave independent creators like Amici behind.”
  • “In Amici’s experience, the tools at issue have created paying customers for human-authored works rather than displaced them.”
  • “For the reasons set forth above, Amici respectfully request that this Court grant Defendants’ motion for summary judgment.”
• ECF 2404

  Reps. Jamie Raskin, Jerrold Nadler and Henry C. “Hank” Johnson Jr.

  Argument leans toward the plaintiffs.

  Declares no sideFiled Oct. 30Proposed brief, 16 pages

  In their words 6 quotes

  “Amici are three senior Members of that Committee. Representative Jamie Raskin is the Ranking Member of the House Judiciary Committee. Representative Jerrold Nadler is a former Chair of the House Judiciary Committee. Representative Henry C. ‘Hank’ Johnson, Jr., is the Ranking Member and former Chair of the IP Subcommittee.”

  • “That is not a specific, tailored exception to copyright law. It’s a bulldozer that would eliminate copyright protection with respect to AI training for all works in all circumstances.”
  • “National security is not a fair use factor. Section 107 of the Copyright Act directs courts to weigh four factors, and national security is not among them.”
  • “The Copyright Office is the federal body with primary expertise in copyright. The Statement appears to have been filed without consulting with that office, which was reportedly surprised by the filing, or even giving it any advance notice.”
  • “Adopting the Statement’s position would be financially devastating to the copyright community and the Nation.”
  • “The Statement fails to appreciate that the financial incentive to create is the means by which copyright achieves its public purpose.”
  • “For the foregoing reasons, the Court should accord the DOJ’s Statement of Interest little if any weight.”
• ECF 2406

  Ken Burns

  Supports plaintiffsFiled Oct. 30Proposed brief, 14 pages

  In their words 6 quotes

  “For nearly fifty years, Ken Burns (‘Burns’) has been making acclaimed documentary films that examine Americans’ shared history and heritage––weaving together complex stories, without taking political sides, to demonstrate how the unum emerges from our pluribus.”

  • “Recently, Burns has learned that several large language models (LLMs), including Defendants’, have copied and ingested dozens (if not all) of his film’s transcripts – and perhaps other copyrighted material that he owns or controls, including companion books to the films – without his permission.”
  • “Here, Defendants’ generative AI models have unilaterally copied and ingested Burns’s works, and are using their contents, without consent, to enable the production of cheaply made facsimiles that are destroying the market for thoughtfully researched, historical documentaries, including Burns’s.”
  • “This negatively reinforcing, doom loop was made possible only by Defendants’ initial, wholesale, unauthorized ingestion of the copyrighted work of Burns and others.”
  • “Here, Defendants’ models create nothing more than verbatim counterfeits of Burns’s works.”
  • “Defendants’ AI products are thus likely to incorporate Burns’s copyrighted content into countless different, infringing outputs, for so long as such products remain available to end users.”
  • “By rendering judgment in favor of the Plaintiffs in this case, the Court would send a clear message to tech companies that have repeatedly flouted the rights of others, for their own financial gain, and provide a meaningful path for content creators to enforce the copyrights in their life’s work.”
• ECF 2407

  Electronic Frontier Foundation

  Supports defendantsFiled Oct. 30Proposed brief, 27 pages

  In their words 6 quotes

  “The Electronic Frontier Foundation (‘EFF’) is a member-supported, nonprofit civil liberties organization that has worked for more than 30 years to protect free expression, access to knowledge, and innovation.”

  • “Plaintiffs are asking the Court to throw that caution to the winds and rewrite copyright law to include a lawful acquisition requirement that finds no basis in statute or case law; and to tilt the fair use analysis against the developer of a general-purpose tool if that tool might be used to create potentially competing non-infringing works.”
  • “In the 21st century, pegging fair use to a demonstration of lawful provenance would cast legal doubt on the billions of copies we collectively make every day through routine online activities like opening emails, clicking on an embedded link, visiting websites, sharing pictures, streaming music, and watching videos.”
  • “In fact, good faith has no significant role in modern fair use analysis.”
  • “That is why copyright law is designed to inhibit infringement—i.e., the proliferation of identical or substantially similar works—not competition.”
  • “And even assuming it were possible to find all of the right licensing partners, depriving training of fair use protection would still ensure that technological development stays in the hands of the very few companies that can afford to make those deals.”
  • “Accordingly, Amicus urges the Court to grant Defendants’ motions for summary judgment.”
• ECF 2408

  Center for AI and Digital Policy, AI experts and legal scholars

  Argument leans toward the plaintiffs.

  Declares no sideFiled Oct. 30Proposed brief, 37 pages

  In their words 6 quotes

  “The Center for AI and Digital Policy (‘CAIDP’) is an independent nonprofit research and educational organization that promotes artificial intelligence policies grounded in fundamental rights, democratic institutions, and the rule of law.”

  • “CAIDP addresses a narrower question than the ultimate fair-use determination. The Court should not interpret copyright law in a way that gives developers an advantage from failing to know, preserve, or disclose facts about training-data provenance.”
  • “The Court should likewise distinguish among acquisition, dataset construction, model training, retention, and outputs rather than treating ‘AI training’ as a single use.”
  • “The point is narrower: when a party’s own evidence identifies the source of training material, the Court may consider any legally relevant characteristics of that source under the proper statutory factor.”
  • “The narrower principle is that courts should not convert an absence of documentation into affirmative support for fair use.”
  • “Technological scale does not eliminate responsibility. Inadequate recordkeeping should not create a legal advantage.”
  • “The Court should apply § 107 to the specific acts and record before it. It should not adopt a categorical rule that all generative-AI training is fair use or that all generative-AI training is infringing.”
• ECF 2413

  PEN America, Protect Democracy Project and Reporters Without Borders (RSF USA)

  Supports plaintiffsFiled Oct. 30Proposed brief, 28 pages

  In their words 6 quotes

  “Amici curiae PEN America, Protect Democracy Project, and Reporters Without Borders, Inc. (RSF USA) are organizations that advocate the First Amendment-based rights of a free press.”

  • “Since the Founding, press freedom and copyright law have worked in tandem to support republican self-government. Defendants’ legal arguments would invert this relationship, jeopardizing the economic viability of the news media under the banner of fair use.”
  • “Yet the fact that Defendants’ assertions of fair use would generate profound economic friction between copyright law and the news media provides independent constitutional reason to reject them.”
  • “Such an economic death cycle is entirely plausible. Increased public reliance on AI has decreased public resort to traditional ‘search,’ reducing click-through rates to the News Plaintiffs’ websites.”
  • “AI substitutes threaten to replace the News Plaintiffs’ journalistic work on an economic level and yet cannot possibly replace them on a democratic one.”
  • “... the facts on this record, alongside the distinct constitutional role played by the News Plaintiffs, support a decision against fair use that need not weaken fair use more generally.”
  • “For the foregoing reasons, the Court should grant the News Plaintiffs’ motion for summary judgment and deny the Defendants’ motions for summary judgment.”
• ECF 2417

  Human Consent Foundation, Creative Artists Agency, Cate Blanchett, Steven Soderbergh and others

  Its supporting memorandum (ECF 2430) says the amici “take no position on the ultimate disposition” of the summary judgment motions; the argument leans toward the plaintiffs.

  Declares no sideFiled Oct. 30Proposed brief, 31 pages

  In their words 5 quotes

  “Human Consent Foundation (‘HCF’), Creative Artists Agency, LLC (‘CAA’), Music Artists Coalition (‘MAC’), National Association of Voice Actors (‘NAVA’), Writer, Inc. (‘WRITER’), The Creators Coalition on AI (‘CCAI’), Cate Blanchett, Nick Hexum and Steven Soderbergh (collectively, the ‘Amici’), respectfully submit this brief to help, support, and assist the Court by correcting the factual record regarding the operational realities of the AI data licensing market ...”

  • “The creative industries already have in place a proven infrastructure and workflows to clear rights for dispersed works at scale and cost-effectively.”
  • “Treating the mere failure to implement a technical tag as an ‘implied license’ inverts both web architecture and copyright law.”
  • “If the market can automate clearance for fragmented personal rights, the transaction-cost barriers for copyrighted works are equally surmountable.”
  • “When a market is established and actively growing, it cannot be dismissed as speculative or infeasible.”
  • “The fact that some rights holders will always say no does not provide a legal basis to eliminate permissions and consents altogether.”
• ECF 2420

  Copyright law professors (Robert Brauneis and 20 others)

  Supports plaintiffsFiled Oct. 30Proposed brief, 29 pages

  In their words 6 quotes

  “Amici, listed in the Appendix, are professors who teach and write about copyright law and related subjects.”

  • “Copyright law permits the free reuse of facts and ideas once a work is lawfully acquired, but it has never permitted the wholesale, unauthorized copying of expressive works to gain access to underlying knowledge.”
  • “The technological novelty of generative AI does not render its operational purpose transformative. The dominant original purpose of plaintiffs’ works and the purpose of defendants’ training use is the same: imparting knowledge, skills, and expressive capabilities.”
  • “When one unauthorized copy of a work is used to build a model that is serving a billion people, there is no doubt that the making of that copy could have a substantial impact on the market value of that work.”
  • “Unlike permissible intermediate copying—which extracts purely functional software elements to facilitate independent human creation and interoperability—defendants harvest human expression itself to train automated systems designed to outperform human authors.”
  • “Defendants essentially ask this Court to create an expansive fair use exemption for generative AI developers. Doing so would establish a dangerous form of AI exceptionalism, stripping human creators of their statutory rights and frustrating copyright law’s constitutional purpose.”
  • “Amici copyright law professors respectfully submit that this Court should preserve the integrity of copyright law by granting plaintiffs’ motion for summary judgment and denying defendants’ cross-motion.”
• ECF 2421

  AI Safety Asia, Future of Life Institute, The Tech Oversight Project, Progressive Democrats of America and 12 other organizations and individuals

  Supports plaintiffsFiled Oct. 30Proposed brief, 30 pages

  In their words 6 quotes

  “Amici curiae are organizations and individuals dedicated to promoting the rule of law and the sound administration of justice.”

  • “The companies are arguing for generative artificial intelligence (‘GenAI’) exceptionalism, and if Defendants’ use is fair use under any theory they or their amici have suggested, then nearly every use by nearly every human is also fair use because the arguments can universally apply to every individual.”
  • “The speculative benefits of what AI ‘might,’ ‘will,’ or ‘could’ do cannot justify present-tense mass infringement.”
  • “But each fair use factor can also be tested against a simpler principle: AI companies should not receive privileges that courts would refuse to extend to humans absent strong justification.”
  • “No Defendant's AI system can extract the facts and ideas from the vast majority of the content at issue without first copying the copyrighted work, including all of its expressive elements.”
  • “Publicly available is not the same as public domain; works aren't in the public domain by default, and they are not fair game to copy and use absent explicit notice from the copyright holder.”
  • “For these reasons, the court should refuse to grant Defendants exceptional treatment under the law.”
• ECF 2429

  Computer & Communications Industry Association, NetChoice and TechNet

  Supports defendantsFiled Oct. 30Proposed brief, 26 pages

  In their words 6 quotes

  • “Generative AI models are mere tools—built by human developers and deployed by human users who supply the initiative, intent, judgment, and editorial selection that drive the creative process.”
  • “The copyrighted works are not copied to be consumed or distributed; they are analyzed to extract statistical patterns that enable a tool to generate new text or other output at the direction of a human user.”
  • “The fourth fair use factor—harm to the market for the copyrighted work—has never been understood to reach competitive pressure from works that are not substantially similar to the copyrighted work and therefore do not substitute for it.”
  • “But accurate information about current events is precisely what copyright leaves free ...”
  • “That some developers have paid for those things does not convert every unlicensed act of statistical learning into the usurpation of a protected market.”
  • “The Court should hold that training a generative AI model on copyrighted works is a fair use.”
• ECF 2434

  Michael Moritz

  Supports neither partyFiled Oct. 30Proposed brief, 13 pages

  In their words 6 quotes

  “Amicus curiae Michael Moritz is a technology investor, philanthropist, author, and former journalist.”

  • “The development of a reasonable, sustainable licensing regime is necessary for the long-term survival of both LLM technologies and independent journalism.”
  • “When users are able to view journalists’ works through the AI-generated answer instead of visiting the publisher’s site, the publisher’s business model is directly undermined.”
  • “LLMs require a steady supply of timely, original, and professionally verified information to survive.”
  • “These agreements demonstrate that LLM developers and publishers can enter into collaborative commercial arrangements that preserve the publishers’ underlying rights to their content and help both the publishers and developers enrich their products.”
  • “The companies developing the LLMs possess the ability to build systems wherein journalists are paid for providing access to their original content, but as the Napster example demonstrates, these arrangements will only develop if the law makes clear that the developers cannot copy the publishers’ content without paying.”
  • “... not take a specific position with respect to the summary judgment briefings filed by the parties, he asks that the court find that LLM developers cannot copy the works of journalists and offer substitutive news without offering any form of compensation.”
• ECF 2437

  Khosla Ventures

  Supports defendantsFiled Oct. 30Proposed brief, 29 pages

  In their words 6 quotes

  “Khosla Ventures is a venture-capital firm based in Menlo Park, California, founded in 2004 by Vinod Khosla, co-founder of Sun Microsystems.”

  • “OpenAI used enormous volumes of long-form text to extract unprotectable facts and ideas, linguistic rules, and statistical patterns in human language to build general-purpose AI models whose downstream uses bear no resemblance to any original work’s expressive content.”
  • “A rule preserving fair use for AI training is therefore essential both to the developers of AI models and to the far larger downstream ecosystem of startups, researchers, and applied AI companies whose work depends on continued access to those models.”
  • “It would harm competition by requiring every AI developer to negotiate individual licenses with thousands or millions of copyright holders — an exercise practically impossible for anyone other than maybe the largest firms and the most sophisticated rightsholders ...”
  • “A ruling that AI training is not fair use would almost certainly have a chilling effect on investment necessary for innovation.”
  • “Far from incentivizing creativity, such a ruling would entrench dominant incumbents at the expense of innovative startups, chill investment, and risk ceding American leadership in AI to foreign jurisdictions with more permissive legal frameworks.”
  • “Khosla Ventures respectfully urges the Court to grant OpenAI’s Motions for Summary Judgment.”
• ECF 2438

  Joseph Stiglitz

  Supports plaintiffsFiled Oct. 30Proposed brief, 51 pages

  In their words 6 quotes

  “Professor Joseph Stiglitz (‘Stiglitz’) is a University Professor at Columbia University with joint appointments in the Faculty of Arts and Sciences (Department of Economics), the Graduate School of Business, and the School of International and Public Affairs.”

  • “In short, this analysis shows that when AI intermediaries absorb the attention and revenue that once flowed to producers of reliable information, the incentive to produce that reliable information erodes, and the stock of accurate information available to society may shrink over time, even if the technology of processing and transmitting information improves.”
  • “If this practice becomes the norm, it could push the information ecosystem past a tipping point to ‘information collapse,’ in which reliable information would not be produced because the producers could not recoup the cost.” (‘information collapse’ is a term from Stiglitz’s own research)
  • “A use whose efficiency gains come at the cost of the disappearance of the underlying works is not transformative. It is just a more efficient way of taking.”
  • “The ‘public benefit’ Defendants offer is therefore self-undermining, as their products draw their value from the very works whose production their uncompensated use discourages.” (‘public benefit’ refers to the defendants’ argument)
  • “This Court should not permit the fair use doctrine to excuse the AI companies’ appropriation. Rather, content owners and AI companies should negotiate licenses to equitably split the benefits from this innovation and the content upon which it depends.”
  • “This Court should bear these dynamics in mind when considering the fair use doctrine to the conduct at issue in this case.”
• ECF 2447

  Computer scientists Mike Knoop, Daniel Lowd and Douglas C. Schmidt

  Supports defendantsFiled Oct. 30Proposed brief, 160 pages

  In their words 6 quotes

  “Mike Knoop, Professor Daniel Lowd, and Professor Douglas C. Schmidt (collectively, the ‘Computer Scientist amici’) respectfully submit this brief as amici curiae in support of defendants’ motion for summary judgment on the defense of fair use.”

  • “This brief explains a narrower technical point relevant to fair use: training is a computational use of text to fit numerical model parameters, not ordinarily a system for storing and retrieving the source works.”
  • “A trained model is not an addressable document repository and does not ordinarily retrieve identified training data as stored files.”
  • “LLMs can sometimes ‘memorize’ training material—meaning generate a nearly exact replica of text they were trained on—such as text that appears numerous times in the training corpus. But LLMs do not memorize most of the text they are trained on, and how much and what a given model memorizes is an empirical question that depends on the particular model and how it was trained.”
  • “In sum, extracting tokens and statistical probabilities from copyrighted works in order to produce a general-purpose tool that can generate text for new prompts and tasks is plainly a ‘further purpose’ with a fundamentally ‘different character’ than the works’ expressive purpose.” (quoting the Supreme Court in Warhol)
  • “Training does not select for a work’s creative elements; it operates over all of the text, and the more commonplace an element is, the more it influences the resulting parameters.”
  • “Amici respectfully submit these scientific observations to assist the Court in applying the relevant legal standards to its determination of fair use.”
• ECF 2448

  America’s Newspapers

  Supports plaintiffsFiled Oct. 30Proposed brief, 25 pages

  In their words 6 quotes

  “Amicus curiae America’s Newspapers is a nonprofit trade association of more than 1,500 local news publishers and related companies across the United States.”

  • “But local news now faces an existential threat at the hands of AI companies that have built some of the most valuable businesses in history using the creative work of local journalists without permission or payment.”
  • “The Copyright Act and the Digital Millennium Copyright Act exist precisely to protect against these abuses. No commercial enterprise may copy protected works wholesale and exploit them in a manner that destroys the market for the originals.”
  • “That business model leaves local publishers acutely vulnerable to any disruption that diverts readers from their websites, which is what Defendants specifically built their products to do.”
  • “If the wholesale copying at issue here is fair use, local publishers will be left in an impossible position: unable to stop AI companies from using their work and unable to demand payment when they do.”
  • “Whatever public benefits Defendants claim their products offer, they cannot outweigh the cost of depriving local communities across the country of their most trusted and essential news source.”
  • “For the foregoing reasons, amicus curiae America’s Newspapers respectfully urges the Court to grant the News Plaintiffs’ motion for partial summary judgment and deny the Defendants’ motions for summary judgment.”
• ECF 2449

  Gregory Leppert

  Supports defendantsFiled Oct. 30Proposed brief, 20 pages

  In their words 6 quotes

  “Gregory Leppert is the Executive Director of the Institutional Data Initiative (IDI) at the Harvard Law School Library, a research initiative that works with libraries, universities, and governments to refine and publish their collections as datasets for computational uses.”

  • “Mr. Leppert takes no position on whether any particular AI output would infringe any copyrighted work, or would be fair use.”
  • “General-purpose AI models, trained on a vast, beautifully variegated corpus of data, offer progress towards a future with libraries in it. Mr. Leppert respectfully submits that the Court should consider this among the relevant public benefits in applying the fair use factors.”
  • “Based on Mr. Leppert’s work, he estimates that, at current rates, it will take libraries hundreds of years to digitize their existing materials, let alone new items added in the meantime.”
  • “A robust ecosystem of general-purpose models is thus integral to libraries’ needs; limiting model training to a subset of materials—curated perhaps by cost, or commercial importance, or the practical availability of a licensing contract—would reduce the effectiveness of those models in preserving and making available materials outside of that distribution.”
  • “In short, Americans who wish to read Game of Thrones do not have to devote hours to piecing it back together using AI. They need only visit their local library.”
  • “Mr. Leppert hopes that the Court will find this discussion helpful as it considers how to resolve the pending motions in this action.”
• ECF 2451

  Institute for Intellectual Property and Social Justice

  Supports neither partyFiled Oct. 30Proposed brief, 9 pages

  In their words 6 quotes

  “The Institute for Intellectual Property and Social Justice (‘IIPSJ’) promotes social justice in the field of intellectual-property law and practice, including copyright, both domestically and globally.”

  • “Moreover, these ‘extra-economic’ interests of authors and the public can be especially acute in marginalized communities, wherein authors and users alike often lack access to both the legal protections and the knowledge, library, and cultural resources available to mainstream communities.”
  • “This ‘hallucination’ problem does a disservice to the user and to the authors whose works are accurate but incorrectly applied by the statistical correlations embedded in the neural network at the core of these AI models.”
  • “If works are not allowed to be used to train the AI models, then the curators of the training materials control the direction of the training and the content on which the AI model is trained. This can result in the voices of marginalized communities being minimized or excluded or of less popular views not being included in the training of the models.”
  • “In light of this history of copyright inequities, the question of incorporation of AI into the copyright ecosystem presents a Gordian Knot of potential boons and perils for marginalized communities ...”
  • “The balancing necessary to develop a just approach might best be done by Congress, however, where the courts address these issues, IIPSJ submits that the IP-social justice sensitive approach is the best one to take into account all dimensions of the issues.”
  • “For the foregoing reasons, IIPSJ respectfully submits that the application and weighing of the fair use factors to the present issues with due consideration for copyright social justice, will best serve the public’s interest in a vibrant and efficacious copyright ecosystem.”
• ECF 2452

  Jim Albrecht

  Supports plaintiffsFiled Oct. 30Proposed brief, 19 pages

  In their words 5 quotes

  “Amicus Jim Albrecht served as Senior Director of News Ecosystem Products at Google from 2017 to 2023.”

  • “The end user never sees those copies because the LLM rewrites them in the answer it provides to the user. But the copies are made nevertheless ...”
  • “How can it seriously be said that content owners objectively intend for LLMs to copy large chunks of their websites to no economic advantage of the content owner, indeed, likely at an economic disadvantage to the content owner?”
  • “In other words, there is no necessity driving OpenAI to take web content and make endless copies. That is just the easiest and cheapest way to get it.”
  • “In short, no reasonable person can use ChatGPT or Microsoft Copilot and fail to see its capacity to replace news products, at least for some significant part of its audience.”
  • “OpenAI wants to have it both ways: the computer re-writing is fair use, just as if a human did it; but the computer’s unauthorized copying and distribution to these virtual humans is an internal computer operation we can ignore ...”
• ECF 2455

  WME (William Morris Endeavor Entertainment)

  Supports plaintiffsFiled Oct. 30Proposed brief, 18 pages

  In their words 6 quotes

  “WME, as one of the nation’s preeminent talent agencies, represents actors, screenwriters, directors, musicians, authors, athletes, digital creators, and public figures in a variety of commercial negotiations to leverage the full potential of their intellectual property, diversify and grow their businesses, and bring projects to life that define the cultural landscape.”

  • “WME clients are losing paid work to AI output built on their own copyrighted works, and they are losing the licensing fees they would otherwise command for the very use Defendants made of those works for free.”
  • “Compensating artists for AI use of their copyrighted material is not a litigation-driven theory; it reflects an economic reality already taking shape.”
  • “The market for AI training licenses clears that bar with room to spare. The News Plaintiffs did not invent it for litigation.”
  • “WME’s experience is that the availability of unlicensed online material makes it harder for rights holders to obtain a meaningful negotiation over the same content. The absence of closed deals does not mean the content was worthless. It can mean the buyers had already helped themselves.”
  • “A rule that excuses unlicensed copying would hand a lasting advantage to developers that took rather than paid, at the expense of those that did the opposite.”
  • “For these reasons, and those stated by the News Plaintiffs, the Court should hold that the fourth factor weighs against fair use and deny Defendants’ summary-judgment motions on that defense.”
• ECF 2457

  Rebuild Local News, LION Publishers, NewsGuild-CWA and five other local news and journalism groups

  Supports plaintiffsFiled Oct. 30Proposed brief, 19 pages

  In their words 6 quotes

  “Rebuild Local News is a nonpartisan nonprofit that develops public policies to respond to the local news crisis.”

  • “This case will determine whether AI companies’ current approach of not compensating news organizations for scraping content will undercut local news outlets’ business models and remove the economic incentives for paying journalists to produce truthful information for our communities.”
  • “As described below, amici believe that interpreting fair use to enable the current AI business model to continue will undermine the local news market in several ways.”
  • “The civic and economic health of communities and of democracy rests on residents having reliable information of the kind traditionally gathered by local journalists, an essential civic function for which social media and AI have been inferior substitutes.”
  • “Amici clearly disagree with the U.S. Justice Department’s statement that ruling in favor of the New York Times would be bad for smaller news outlets. Amici argue that the opposite is true.”
  • “If this Court were to reaffirm strong copyright protection by ruling for the New York Times, then small media would have a much stronger claim for compensation to fund the development of critically important news media.”
  • “Amici urge the Court to reject the fair use defenses for AI scraping and ingestion of our creations.”
• ECF 2459

  Former Sen. Rick Santorum and former Rep. Rick Boucher

  Supports defendantsFiled Oct. 30Proposed brief, 24 pages

  In their words 6 quotes

  “Amici are former Members of Congress from different parties who share a longstanding commitment to the goal of securing meaningful incentives for authors while preserving room for innovation that has made the United States the world’s leader in emerging technologies.”

  • “The use of copyrighted works to train AI models is a paradigmatic case of fair use. An AI model relies on works not to communicate their expression to the public but rather, quite differently, to derive the statistical patterns of language that allow the model to function.”
  • “AI training in itself does not make anything available to the public, nor do the statistical relationships a model derives have any necessary connection to the plagiarism of copyrighted content.”
  • “Specifically, universities, scientists, nonprofits, libraries, and smaller developers will be harmed by the imposition of a de facto licensing regime.”
  • “Finally, finding fair use here is crucial to national security. The United States and China are locked in a race to develop increasingly sophisticated AI models.”
  • “Any change to that framework is for Congress to make, not the federal courts.”
  • “For the foregoing reasons, the Court should hold that the use of copyrighted works to train AI models is a fair use under 17 U.S.C. § 107.”
• ECF 2460

  AI Startups (BandLab, Boomy, Created by Humans, Credtent, Splice, Newstex, ProRata, SourceAudio, Cashmere, Voice-Swap)

  Supports plaintiffsFiled Oct. 30Proposed brief, 22 pages

  In their words 5 quotes

  “Amici are startup companies at the forefront of AI that respect intellectual-property rights. Each amicus builds AI products from licensed content or runs a platform through which that content is licensed.”

  • “Their experience shows that copyright protection fosters more innovation, not less, for both AI developers and creators.”
  • “No dire consequences will ensue if AI developers must get copyright owners’ permission before training models on their work.”
  • “If AI companies can poach those sources’ work without paying for it, the diversion of traffic will threaten some companies’ economic survival.”
  • “Requiring AI developers to get permission before training models on copyrighted data will change what gets built and how it gets built in a way that benefits all stakeholders; it will not slow innovation.”
  • “Licensing has turned out to be pro-innovation and profitable for all sides, protecting existing markets while creating new ones ...”
• ECF 2461

  Midjourney and Runway

  Supports defendantsFiled Oct. 30Proposed brief, 35 pages

  In their words 6 quotes

  “Midjourney, Inc. (‘Midjourney’) is a small artificial intelligence (‘AI’) research lab dedicated to creating tools that augment and expand the human imagination. Runway AI, Inc. (‘Runway’) is an applied AI research lab building foundational real-world AI models, including video generation, world simulation, and robotics models.”

  • “Generative AI models are perhaps the greatest engine of new expression and innovation that has ever been devised, and their training on the world’s information, writings, and images to provide these capabilities benefits the public and is quintessential fair use.”
  • “It is not possible to license the volume and diversity of content necessary to train a model capable of creating transformative works like these.”
  • “Nor would licensing be required even if such a market could develop.”
  • “AI thus functions as a complement and tool to augment human creativity, not a substitute for it.”
  • “Without a finding of fair use here, these innovative developments and many we haven’t even thought of yet risk being stopped in their tracks.”
  • “Their training should be recognized for what it is: a quintessential transformative fair use that serves the core purposes of copyright by enriching our common culture and advancing the arts and sciences for the public benefit.”
• ECF 2462

  Trevor Lawrence Jr., Chamber of Progress and Abundance Institute

  Supports defendantsFiled Oct. 30Proposed brief, 20 pages

  In their words 6 quotes

  “Amici are a multi-Grammy-winning musician and mission-driven nonprofit organizations committed to ensuring that interpretations of copyright law promote genuine creativity ...”

  • “The training use at issue puts the works to a fundamentally different purpose: extracting non-expressive, linguistic patterns as the basis for general-purpose, predictive language models. This is a paradigmatic transformative use, strongly favoring fair use.”
  • “Copyright protects an author’s particular expression of ideas, but does not extend to analytical information about that expression.”
  • “In each instance, generative AI broadens the musical possibilities available for an artist to consider while leaving composition, selection, and performance to human judgment.”
  • “Requiring permission for the training use at issue, as Plaintiffs’ position would require, would materially burden the development of tools that support new creative work.”
  • “Burdening those opportunities through the permission requirement that Plaintiffs seek would leave the public with fewer new works and creators with fewer paths to producing them.”
  • “For the foregoing reasons, the Court should deny Plaintiffs’ motions for summary judgment insofar as they challenge the training uses addressed here and grant OpenAI’s motions for summary judgment on fair use as to those uses.”
• ECF 2465

  Teodor James Holmberg

  Filed Nov. 2, after the Oct. 30 deadline.

  Supports plaintiffsFiled Nov. 2Proposed brief, 27 pages

  In their words 6 quotes

  • “Accordingly, any formatting, modification, and/or preparation of copyrighted works to make them suitable for AI input may be considered the creation of a derivative work.”
  • “Just because copyrighted materials are the indispensable fuel required by the AI engine does not make its use a fair one.”
  • “While the AI output may add something new and be transformative, what the defendants want and need for AI input is the original expressions of the plaintiffs’ copyrighted works.”
  • “However, AI technology requires the use of copyrighted material as input and, as such, the use of copyrighted works as AI input should be recognized as a protectible property interest with its own market.”
  • “In other words, the ‘hot news’ exception is an example of a possible misappropriation claim under state law, not the only type of un-preempted misappropriation claim under state law.”
  • “I respectfully request that the Court consider the two different, mutually exclusive alternatives described herein: (i) the use/market for copyrighted works as AI input is a derivative use/market under 17 U.S.C. §106(2), to which the doctrine of fair use does not apply; or (ii) the use/market for copyrighted works as AI input is a new use/market not covered by the copyright statute and thus is a use/right which may be the basis for a state law/common law cause of action not preempted under 17 U.S.C. §301.”
• ECF 2472

  E.H. and W.E. (high school students, pro se)

  Supports plaintiffsFiled Oct. 30Proposed brief, 10 pages

  In their words 6 quotes

  “Amici Curiae E.H and W.E are high school students. We are not corporation owners, litigation funders, or industry lobbyists.”

  • “But from our perspective, it is not transformative; it is a replacement.”
  • “Generative AI does not engage with that expression. It does not critique it or parody it. It simply ingests it to mimic it.”
  • “'Web grounding' doesn't just summarize people's work; it replaces the need for the consumer to visit the original site.” (‘Web grounding’ is Microsoft’s term)
  • “This means that the photos, essays, and social media posts we created when we were thirteen have likely been scraped and used to train AI.”
  • “We are the first generation to grow up with the digital tools to share our creativity globally, only to be told we must abandon those tools because the digital space has been flooded with unlicensed AI.”
  • “For the foregoing reasons, Amici Curiae respectfully requests that the Court reject Defendants’ fair use defense, and give full weight to the documented shrinking of the new creator market, ruling in favor of Plaintiffs.”
• ECF 2482

  Muhammad Deckri Algamar

  Filed Oct. 31, after the Oct. 30 deadline.

  Supports neither partyFiled Oct. 31Proposed brief, 12 pages

  In their words 5 quotes

  “Muhammad Deckri Algamar, AIGP, is an LL.M. candidate at IE University in Madrid who researches copyright and AI governance.”

  • “This brief takes no position on whether training an AI model on a book is fair use. It addresses a question that survives either answer: how does an author find out whether a book was used?”
  • “An author should not need a federal lawsuit to learn whether a book went into a training set.”
  • “A developer with a fair use to defend loses little by saying which books it used.”
  • “The yes-or-no check is the least intrusive, since it reveals nothing beyond one work the author already knows about.”
  • “The title of a published book and the name of a dataset are different in kind. Amicus asks that any request to seal them meet the requirement of specific, on-the-record findings.”
• ECF 2497

  Professors Jon M. Garon, Adam Mossoff, Loren E. Mulraine, Chris Newman, Sean Pager, Zvi Rosen and Mark F. Schultz

  Supports plaintiffsFiled Nov. 6Proposed brief, 23 pages

  In their words 6 quotes

  “Amici are scholars and professors whose research focuses on copyright law.”

  • “Defendants seek an AI-training exemption that Congress never enacted, the Constitution does not recognize, and Supreme Court precedent forecloses. Calling the copying ‘training’ does not excuse the unauthorized reproduction of thousands of copyrighted books.”
  • “A robust licensing market sustains incentives for human authorship while supplying the high-quality inputs AI models require. If widespread unauthorized copying is deemed fair use, that market may disappear.”
  • “Obtaining the books, retaining or sharing copies, using them to develop a model, and deploying that model are distinct uses, each requiring separate analysis.”
  • “But a model trained on novels to generate novels or story-like prose uses the source works to acquire capabilities valuable in the same market.”
  • “Beyond licensing markets, AI outputs generated by models built on copyrighted works dilute the market for those works—a distinct harm cognizable under factor four.”
  • “For these reasons, amici curiae Professors Jon M. Garon, Adam Mossoff, Loren E. Mulraine, Chris Newman, Sean Pager, Zvi Rosen and Mark F. Schultz respectfully request that the Court (1) deny Defendants’ motion to the extent that it seeks a categorical ruling that copying Plaintiffs’ books for model training, research, or development was fair use; (2) grant Plaintiffs’ motion on the distinct unauthorized downloading, copying, and distribution uses to the extent that the record is undisputed; and (3) apply Section 107 to the particular uses and facts in the record.”