DOJ Backs OpenAI: AI Training Is Fair Use

The Justice Department told a Manhattan federal court this week that training large language models on copyrighted material is generally fair use, siding with OpenAI against The New York Times and the other publishers suing it. The filing landed Tuesday before US District Judge Sidney Stein in the Southern District of New York, and it is the first time the federal government has stated a formal position in the wave of copyright suits brought against AI companies by authors, publishers, music labels and news organizations since 2023.

The government’s argument is framed around national competitiveness rather than copyright doctrine. The United States has a “strong interest in continuing to develop a robust and competitive” AI industry, the department wrote, and “the creative possibilities and public benefits” of training language models on such content “far outweigh any competitive harm.” Ruling for the publishers, it warned, would obstruct “creative and scientific progress while hindering American prosperity.” The brief also leans on a national-security rationale — that models capable of analyzing real-world information serve defense and intelligence needs — and characterizes the training itself as transformative, the pivotal question in any fair-use analysis.

Procedurally, the filing is a statement of interest, not a ruling or a party brief. The mechanism lets a federal agency put the government’s views in front of a judge without joining the case, and it carries no binding legal authority; Stein is free to disregard it entirely. It was signed by Associate Attorneys General Stanley Woodward Jr. and Brett Shumate along with senior counsel Michael Weisbuch. What it does carry is weight of a different kind — an appellate court weighing fair use for AI training will now be doing so with the executive branch’s position already on the docket.

The consolidated case in front of Stein has grown well beyond the Times. It now gathers claims from The Intercept, MediaNews Group, Tribune Publishing, Ziff Davis, the Center for Investigative Reporting and individual writers including Sarah Silverman. The Intercept’s own suit dates to February 2024 and took a narrower route than most, pleading a Digital Millennium Copyright Act claim over stripped author and title metadata rather than straight infringement; in November 2024 the court dismissed the claims against Microsoft but let the DMCA claim against OpenAI proceed. That mix matters, because a fair-use holding would not dispose of the metadata theory.

The publishers were blunt. “The Administration is siding with a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole,” a Times spokesperson said, adding that “both AI and creators can thrive — AI companies simply need to pay fairly for the content that makes their products possible, as copyright law requires.” Matt Topic, counsel for The Intercept, said accepting the government’s position would amount to “an unprecedented, uncompensated transfer of IP rights from news organizations to tech companies.”

The timing is the part worth noting. Fair use for AI training has been litigated case by case for two years, with district judges reaching split conclusions on near-identical facts and several defendants choosing to settle rather than test the question on appeal. A statement of interest does not resolve that split, but it changes the negotiating table: publishers weighing whether to settle now have to price in a federal government that has publicly aligned itself with the defense, and AI companies weighing whether to license have one fewer reason to.

Stein has given no indication of when he will rule on the pending motions. Whatever he decides, the case is widely expected to reach the Second Circuit, which would be the first appellate court to squarely address whether ingesting copyrighted journalism to train a commercial language model is transformative — and by then, the executive branch will have been on record for months.