Washington's filing calls model training fair use, the first time the federal government has taken a formal side in the copyright fights reshaping the AI industry.

The U.S. Justice Department has told a Manhattan federal court that training large language models on copyrighted text is fair use, putting the federal government's weight behind OpenAI in its copyright fight with The New York Times. The filing, submitted this week as a statement of interest, is the first time Washington has formally taken a side in the wave of copyright suits that authors, publishers, music labels and news organizations have brought against AI developers. It does not decide anything.

What it signals is where the Trump administration wants the law to land.

What the government filed

A statement of interest lets a federal agency set out its position without joining a case as a party, and it carries advisory rather than binding weight. The Justice Department filed one before U.S. District Judge Sidney Stein, who will still rule on his own reading of the law. Government lawyers argued that language models do not copy the works they train on. In their account the systems use those works to build general language ability, a process the brief describes as extraordinarily transformative under copyright law. The department pushed the argument past this single lawsuit, telling the court that models are already helping researchers make breakthroughs across scientific fields.

The filing frames the stakes in national terms. It ties the outcome to scientific progress and national security, and it leans on an executive order President Trump signed in January 2025 on removing barriers to American leadership in AI. A narrow reading of fair use, the government wrote, would slow research and weaken American prosperity and economic mobility.

Associate Attorney General Stanley Woodward defended the intervention publicly, casting AI leadership as a matter of national security and economic opportunity while adding that the administration also wants to protect artists.

The case it is trying to influence

The New York Times sued OpenAI and its largest backer, Microsoft, in December 2023, accusing them of using millions of its articles without permission to train the models behind ChatGPT. Court records show the case has since absorbed related claims from other newspapers against the same defendants. The Times is seeking billions of dollars and argues that OpenAI's products compete directly with journalism and pull readers away from the source.

The dispute has grown combative. In July 2026 the Times asked the court to sanction OpenAI, accusing the company of concealing evidence tied to how it handled training data during discovery.

OpenAI, for its part, has said for years that it trains on publicly available data and relies on longstanding fair use principles, a position it has published on its own site in response to the suit.

Fair use, and why the question is unsettled

Fair use is the part of copyright law that permits some unlicensed uses of protected work. Courts weigh four factors when they apply it: the purpose and character of the use, the nature of the original work, how much of the work was taken, and the effect of the use on the market for the original. The decisive battle in the AI cases sits inside the first factor, over whether feeding books and articles into a model is transformative enough to qualify.

Judges have not lined up on the answer. Two federal courts reached opposing conclusions in 2025, and no appeals court has resolved the conflict. In the Anthropic matter Judge William Alsup treated training as fair use, while a second San Francisco judge, Vince Chhabria, warned in a parallel suit against Meta that using copyrighted work without permission to train AI would be unlawful in many circumstances. That split is why the Times case reaches past its own docket, and it is the gap the government's brief is trying to fill.

The Anthropic precedent hanging over the fight

The most consequential ruling so far landed in that separate suit against Anthropic, the maker of Claude. In June 2025 Alsup found that training a model on lawfully acquired books was fair use and called the use exceedingly transformative. In the same decision he drew a hard line on how the books were obtained.

Anthropic had downloaded more than seven million books from pirate repositories known as shadow libraries, and Alsup held that assembling a permanent collection of pirated copies was not fair use, whatever the copies were later used for. The company then agreed to pay $1.5 billion, roughly $3,000 for each of about 500,000 works, in what lawyers involved called the largest copyright recovery in U.S. history. A federal judge granted final approval in July 2026, after Alsup had retired and the case passed to Judge Araceli Martinez-Olguin.

Here is the distinction creators keep returning to. Anthropic was not penalized for training on the books. It was penalized for how it got them, which leaves the core question the Times is pressing, whether the training itself is lawful, still open for a court to answer.

What the brief does and does not change

The government cannot decide this case, and the lawyers who signed the brief hold no jurisdiction over it. Their argument enters the record as persuasion, and Judge Stein stays free to reject every word of it.

The alignment itself is not a surprise. OpenAI has argued the fair use position for years and asked the administration to write it into national policy through its AI Action Plan proposal, so the filing backs a stance the company already held. The White House and OpenAI did not respond to requests for comment on the filing, and the Times stayed silent as well.

What is new is the sight of the federal government reaching into a private copyright suit to argue that one side should win. Until now the courts had refereed these fights on their own, and the result still rests with a judge who owes the administration's brief nothing.