OpenAI Touts Briefs Backing Fair Use for AI Training

OpenAI published a post on Friday, titled “Fair use benefits everyone,” featuring excerpts from 20 filings that urge the U.S. District Court for the Southern District of New York (SDNY) to find that training large language models (LLMs) on copyrighted works is fair use. All but one are amicus briefs filed on or before the October 30 deadline in In re OpenAI, Inc. Copyright Infringement Litigation, the multidistrict litigation (MDL) that combines copyright lawsuits filed against OpenAI and Microsoft by The New York Times and other news publishers, Ziff Davis, and a group of authors, led by the Authors Guild, seeking to proceed as a class.

The excerpted filers include library and disability groups, startup and technology advocates, law professors, computer scientists, authors, Khosla Ventures, former Senator Rick Santorum and former Representative Rick Boucher, and the U.S. government. OpenAI wrote that specialists in many disciplines “have recognized that fair use is a win-win that benefits everyone.” The post said the Times’ lawsuit “asks everyone to ignore a half century of legal history and progress” in pursuit of “an undeserved payday.”

Among the excerpts is the Statement of Interest filed on September 1 by the U.S. Department of Justice (DOJ). The DOJ urged the court to hold that using copyrighted works to train LLMs is fair use and cautioned that conditioning training on licenses would favor the few technology firms able to pay, to the detriment of U.S. competitiveness and national security. The DOJ also criticized the 2025 decision in Kadrey v. Meta Platforms, stating that “the Kadrey court’s application of the fourth fair-use factor is deeply flawed.”

Cross-motions for summary judgment on fair use are pending before Judge Sidney H. Stein. Earlier fair use rulings on AI training came from two judges in California, and Judge Stein’s decision will be “the first case to be decided in New York, the first involving OpenAI, and the first involving consolidated claims from both news and book publishers and authors,” according to a memo from Re:Create Executive Director Brandon Butler, whose members are among those who filed in support of fair use.

Electronic Frontier Foundation

The Electronic Frontier Foundation (EFF) argued that the Times and the other rightsholders want the court to “rewrite copyright law to include a lawful acquisition requirement that finds no basis in statute or case law.” EFF also objected to penalizing makers of general-purpose AI tools in the fair use analysis since users might create rival, noninfringing works. EFF contended that treating unauthorized copying as disqualifying would endanger public-interest research and journalism that rely on scraping, along with everyday intermediate copying.

Public Knowledge

Public Knowledge urged the court to analyze training on its own, apart from how the data was obtained and what the models produce, and contended that training is fair use since the models learn uncopyrightable information about how language works without standing in for the source material. The brief cited earlier fair use rulings covering image search engines, Google’s book-scanning project, and library copies made for blind and print-disabled readers. In the view of Public Knowledge, “copyright is not a general remedy for technological disruption or a means for reallocating the gains from a new technology,” and fair use can help spread the gains from AI widely rather than leaving them “controlled by a privileged few.”

Professor Matthew Sag

Emory University School of Law Professor Matthew Sag submitted a brief that does not back either side, though OpenAI included it among the excerpted filings. Sag asked the court to center its analysis on “expressive substitution” and contended that training qualifies as transformative given that, in ordinary use, the models’ outputs “are in no relevant way similar to the invisible intermediate copies made during the training process.” Sag also maintained that no workable market could form for licensing the volume of material needed for training.

The brief flagged two issues; one concerns whether copyright should recognize “indirect expressive substitution,” his term for a system designed around a particular source with the goal of supplanting it, and the other concerns books OpenAI drew from so-called “shadow libraries” in its early training, a matter Sag said the court should consider when weighing market effect under the fourth factor. “My aim is to help the court get the doctrine right, whoever wins on these particular facts,” Sag said in a press release.

Authors Alliance

Authors Alliance limited its brief to whether copyright treats “market dilution” as a legally relevant injury. The brief asked the court to rule that “aggregate competitive pressure from noninfringing outputs” does not qualify as cognizable market harm under Section 107(4). Authors Alliance cautioned that such a rule would fall mostly on scholars, biographers, emerging writers, and small presses, which depend on fair use and cannot afford to defend against broad claims of market harm.

Computer and Communications Industry Association

The Computer and Communications Industry Association (CCIA), which filed jointly with NetChoice and TechNet, described training a generative AI model as “among the most clearly transformative uses a court has been asked to evaluate.” Rather than being made available for reading or distribution, the works serve as material for identifying patterns in data so that a tool can produce new content when a person prompts it. CCIA also dismissed market dilution as “an unrecognized and unrealistic theory of harm,” in its announcement.

Briefs supporting the Times and other rightsholders were also filed, including from the Association of American Publishers, William Morris Endeavor Entertainment, LLC, The Human Consent Foundation, and a brief by Representatives. Jamie Raskin (D-MD), Jerrold Nadler (D-NY), and Hank Johnson (D-GA), which challenged the DOJ’s national security arguments. Those supporting the Times are urging the court to rein in LLMs: “Fair use is not a roving license for courts to revise the framework Congress implemented whenever new technologies emerge and markets are still developing,” said the Association of American Publishers’ brief.

The deadline for the parties’ reply briefs in the case is November 20.