Justice Department tells a federal court that LLM training is fair use

On September 1, 2026, the United States filed a Statement of Interest in In re: OpenAI, Inc. Copyright Infringement Litigation, the consolidated multidistrict case before Judge Sidney Stein in the Southern District of New York (25-md-3143). The government appeared under 28 U.S.C. Section 517, which lets the Justice Department attend to the interests of the United States in any pending federal suit and requires no leave from the court. The filing is signed by Associate Attorney General Stanley Woodward, Assistant Attorney General Brett Shumate, and Senior Counsel Michael Weisbuch.

The nineteen-page brief argues that training a large language model on written works is fair use. It leans on the first and fourth statutory factors, describing training as radically transformative and citing Google v. Oracle, Authors Guild v. Google, Bartz v. Anthropic, and Kadrey v. Meta. It attacks the plaintiffs’ market dilution theory as untethered from copyright’s purpose, arguing that generalized competitive harm from a new technology is not a cognizable market effect and that outputs must be analyzed separately from training. The government also frames the question in national security terms, arguing that rules making a robust domestic AI industry harder to build hand an advantage to foreign adversaries. Notably, the brief says the correct venue for changing the rule is Congress, not the courts: whether the technology warrants a wholesale rewriting of copyright principles is a policy question best left to elected representatives.

This is the first time the federal government has taken a formal position in the wave of copyright suits filed against AI developers, and the brief says its reasoning applies to all parties in the litigation and the related book author and publisher cases. It is not a ruling and it does not resolve how training data was acquired or whether specific outputs infringe. For anyone building on or licensing frontier models, it is a strong signal about which way federal policy is pushing, and a reminder that the acquisition question, which sank Anthropic in the Bartz pirated-library findings, remains separate and live.