Summary
On September 1, 2026, the US Department of Justice filed a 20-page statement of interest in In re OpenAI, Inc. Copyright Infringement Litigation, the multidistrict proceeding before Judge Sidney Stein in the Southern District of New York that consolidates The New York Times Company's suit against OpenAI and Microsoft with related cases brought by authors and publishers. The brief, filed under 28 U.S.C. § 517 and signed by Associate Attorney General Stanley E. Woodward Jr., Assistant Attorney General Brett Shumate and Senior Counsel Michael Weisbuch, asks the court to hold that copying written works to train large language models is fair use and to reject the "market dilution" theory articulated in Kadrey v. Meta. It is the first time the federal government has formally stated a position in the copyright suits against AI developers. A statement of interest is not binding; Stein decides the question independently.
What Happened
The filing appears as docket entry 1682 in the MDL, marked as relating to "All Matters." A footnote says that, "to simplify," the government refers only to the Times and OpenAI, but that its arguments "apply similarly to all parties in this litigation and the related cases, including book authors and publishers." A second footnote disclaims any suggestion that the government authorized the conduct at issue under 28 U.S.C. § 1498. The interest section cites President Trump's January 23, 2025 executive order "Removing Barriers to American Leadership in Artificial Intelligence," his June 2, 2026 order "Promoting Advanced Artificial Intelligence Innovation and Security," the March 2026 legislative recommendations in the administration's National Policy Framework for Artificial Intelligence (which the brief quotes as stating that "training of AI models on copyrighted material," in and of itself, "does not violate copyright laws"), and a 2022 Government Accountability Office report on AI and national security. The brief argues that "rules of law that make it significantly more difficult to develop a robust AI industry in the United States therefore threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered."
The legal argument confines itself to the training stage. Citing Andy Warhol Foundation v. Goldsmith, the government says fair use requires a use-by-use analysis, so the copying of works to build a model must be assessed separately from the acquisition of training data and from any given output. On the first factor it calls training "transformative—spectacularly so," quoting Judge William Alsup's June 2025 ruling in Bartz v. Anthropic, and "extraordinarily transformative"; it likens the use to the code copying approved in Google v. Oracle. On the fourth factor it adopts the Second Circuit's Authors Guild v. Google standard that only "significant substitutive competition" counts, reasoning that "training does not reveal anything to the public at all." A footnote adds that even if some outputs reproduced protected text, "a tiny sliver of anomalous reconstructive outputs would not support a remedy that cuts off or threatens massive liability for LLM output uses generally," citing Trump v. CASA on the scope of remedies.
The brief devotes its longest passage to attacking Judge Vince Chhabria's Kadrey v. Meta opinion, which it characterizes as "contrary dicta" adopted "without the benefit of briefing." It calls Kadrey's indirect-substitution theory of market dilution "deeply flawed" for collapsing training and outputs into a single continuous use and applying a "genre-level understanding of 'substitution.'" To illustrate, it recounts that a teenage Joan Didion typed out Hemingway's stories "to learn how the sentences worked," and says that under Kadrey's logic she would have owed Hemingway for every later piece she published. A further footnote says the Register of Copyrights, "who is currently challenging her removal," endorsed a similar theory in the Copyright Office's May 2025 Part 3 report on generative AI training, and that her view "does not warrant deference" under Loper Bright.
The government also makes policy arguments outside the four factors. It says a licensing requirement would let only the largest technology companies afford training, would "disproportionately benefit legacy media outlets due to the sheer volume of their written publications," and would function "primarily as large subsidies for old mainstream media companies." In a footnote it takes "no position on whether a licensing regime would be financially or logistically feasible," and it notes that publishers can and do license real-time and paywalled content voluntarily. It says the question whether the technology "warrants a wholesale rewriting of copyright principles is a policy question best left to the People and their elected representatives." It cites a report that the Times itself has issued AI-use guidance to freelancers, and an independent writer's LLM-assisted critique of a Times article on data-center water use, as evidence that publications large and small are using the technology.
The filing arrived during a procedural squeeze on the Times case. On August 28 (entered August 31), Stein ordered the Times to show cause by September 11 why its action should not be stayed pending resolution of summary-judgment motions in the other active MDL cases, with defendants to respond by September 18. On September 4 the news plaintiffs (the Times, Daily News and Ziff Davis) and Microsoft moved to seal portions of a combined summary-judgment brief and statement of undisputed facts. Times spokesperson Graham James told Bloomberg Law that "the Administration is siding with a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole," and that AI companies should "pay fairly for the content that makes their products possible, as copyright law requires." Three days after the filing, The Seattle Times and Newsday filed a separate copyright and trademark suit against OpenAI and Microsoft.
Why It Matters
Until now the executive branch's only formal contribution to the training-data question was the Copyright Office's May 2025 report, which the Office issued the day before the Register was removed and which leaned toward a market-dilution reading of the fourth factor. This brief puts the Department of Justice on the opposite side of that report and asks the court presiding over the largest consolidated AI copyright docket to pick the Alsup reading of Bartz over the Chhabria reading of Kadrey. Both of those June 2025 opinions came from the Northern District of California and bind no one in New York; a Second Circuit district court deciding summary judgment in the MDL, with the government's position on the record, is a different posture.
The brief's silence is as instructive as its argument. It confines itself to the training stage and says nothing about acquisition, the stage on which Alsup found Anthropic's use of pirated libraries unprotected and which produced the $1.5 billion Bartz settlement. It also concedes that output-stage questions "may not be transformative" where a model reconstructs a work. How Stein weighs the filing, whether the government will file similar statements in the authors' cases as its footnote implies, and whether the argument survives an eventual appeal are all open. A statement of interest signals the position the United States would take as amicus in any appeal, but it creates no precedent on its own.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.