The U.S. Department of Justice told a federal court that training large language models on copyrighted text generally qualifies as fair use, siding with OpenAI and Microsoft in the sprawling copyright litigation brought by The New York Times and dozens of other publishers and authors.
The government filed its brief, a Statement of Interest, on September 1 in the U.S. District Court for the Southern District of New York, addressing the consolidated In re OpenAI, Inc. Copyright Infringement Litigation (MDL No. 25-md-3143) before Judge Sidney Stein, according to court records and legal trackers who reviewed the filing. Filed under 28 U.S.C. § 517, a Statement of Interest lets a federal agency stake out a legal position without becoming a party to the case — it carries no binding authority, but signals where the administration wants the law to land as the case heads toward summary judgment.
“Transformative — spectacularly so”
According to the filing, copying protected articles to train a model is “a use of a different kind or character” than the original works, because training converts text into statistical and linguistic relationships rather than reproducing it for readers. Citing prior rulings, the DOJ argued the process is “transformative — spectacularly so.”
The department also rejected a “market dilution” theory some rights holders have pressed, under which AI output harms publishers simply by competing in the same genre as human-written work, even without copying specific passages. The DOJ called that reading of fair use’s fourth factor “deeply flawed,” arguing that outputs must be judged separately from training under the Supreme Court’s Warhol precedent, and that dilution claims risk stretching copyright protection over ideas and styles rather than actual expression.
A national-security argument
The brief leans on the administration’s AI policy, including a January 2025 executive order and a March 2026 National Policy Framework for Artificial Intelligence, arguing that requiring licenses for training data would concentrate the large-language-model market among a handful of well-funded companies and weaken U.S. competitiveness. It is the first time the federal government has formally weighed in on the fair-use question running through the OpenAI copyright suits — a question also central to separate litigation against Anthropic, Meta and other labs.
The filing lands in a case that has already turned adversarial over process: publishers have separately asked the court to sanction OpenAI over how it has handled evidence in discovery. The fair-use question itself, including how it differs for AI compared with earlier copying disputes, is explained in more depth in an earlier explainer. Judge Stein has not yet ruled on summary judgment.