DOJ Tells Court AI Training Is Fair Use, Forgets to Mention It’s Negotiating a Stake in OpenAI

The Justice Department has weighed in on the biggest copyright fight in artificial intelligence, telling a federal judge that training a large language model on copyrighted text is fair use. What the DOJ’s filing didn’t mention: the same administration has spent the better part of a year in talks with OpenAI about taking a direct equity stake in the company whose legal position it just endorsed.

What the DOJ Actually Filed

On September 1, the Department of Justice filed a Statement of Interest with the U.S. District Court for the Southern District of New York, urging the court to rule that using copyrighted written works to train large language models qualifies as fair use. The filing landed in the consolidated multidistrict litigation against OpenAI, which folds together The New York Times’ original December 2023 lawsuit with a wave of related claims from other publishers and book authors, all overseen by U.S. District Judge Sidney Stein.

It’s the first time the federal government has formally staked out a position in the string of copyright lawsuits authors, musicians, and news organizations have filed against AI companies over their training data. A statement of interest doesn’t bind the court, but it puts the government’s legal weight behind one side of the argument.

The brief leans on two of the four fair-use factors: the purpose and character of the use, and its effect on the market for the original work. On the first, the DOJ argues that training is “exceedingly transformative,” since a model doesn’t read an article the way a human reader does — it converts the text into statistical relationships about vocabulary and syntax, which is a fundamentally different function than the one the article was written for. On market harm, the government argues that training alone doesn’t put a copy of an article into public circulation and therefore doesn’t compete with it in the way copyright law is meant to police, even if a chatbot’s output later competes with news outlets in a general sense.

The filing also took direct aim at a 2025 ruling in a related case, Kadrey v. Meta Platforms, where a judge suggested that training copies of copyrighted works would usually be found infringing. The DOJ called that court’s reasoning “deeply flawed,” arguing it improperly blended the training step and the output step into one continuous act and treated ordinary market competition as the kind of harm copyright law actually protects against.

The government did carve out some limits. It said it wasn’t taking a position on whether OpenAI legally acquired the training data in the first place, and it acknowledged that a model spitting back near-identical copies of a work on the output side could raise separate legal problems — its argument was limited to the training process itself.

The National Security Framing

Much of the brief’s persuasive weight rests on economic and national-security stakes rather than pure copyright doctrine. The government argued that requiring AI developers to license every copyrighted work used in training would create a cost barrier only the largest tech companies could clear, entrenching a handful of dominant players and disproportionately rewarding media organizations that happen to hold large archives. It framed unrestricted AI development as tied to U.S. competitiveness against rival nations and cited a White House cybersecurity order and CISA directives to argue that AI’s role in cybersecurity gives the government a stake in how the cases come out.

Critics of the brief have pointed out that the case at the center of it began with a far less sweeping dispute — comedians suing over OpenAI’s use of their stand-up sets, and The New York Times objecting to its journalism being used without compensation — and questioned whether that really rises to the level of a national security matter.

The Stake Talks Nobody Mentioned in the Brief

What the DOJ’s filing doesn’t reference anywhere is a separate, ongoing negotiation between the Trump administration and OpenAI over the government taking a direct financial stake in the company. Talks on the idea have been running for more than a year, according to OpenAI, and became public in June when President Trump confirmed discussions aboard Air Force One, describing an arrangement where “the American public essentially becomes a partner.” By July, the Financial Times reported that OpenAI CEO Sam Altman had proposed the government take a 5% equity stake — worth roughly $42.6 billion at OpenAI’s then-$852 billion valuation — funneled into a proposed sovereign-wealth-style “Public Wealth Fund” that would let ordinary Americans share in AI’s financial upside. The talks reportedly involved Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, and Altman floated the idea to other major AI labs as well, though none have confirmed similar arrangements.

Nothing about the equity talks has been finalized, and OpenAI has publicly cast the idea as a way to share AI wealth with the public rather than as compensation for anything specific. But the optics are hard to ignore: an administration that may soon hold a multibillion-dollar ownership stake in OpenAI has just filed a brief that would spare the company from potentially massive copyright liability, without disclosing the financial talks anywhere in the filing. The government has precedent for exactly this kind of arrangement — it took a 9.9% stake in Intel in 2025 tied to CHIPS Act funding, and has since taken equity positions in several rare-earth and quantum computing firms — but those deals didn’t coincide with the same agency simultaneously arguing a legal position that benefits the company in active litigation.

Why It Matters Beyond OpenAI

The DOJ’s position, if adopted by the court, wouldn’t just resolve the Times’ case. The brief explicitly said its reasoning extends to every related claim folded into the multidistrict litigation, covering authors, publishers, and other copyright holders who’ve sued over AI training practices. A ruling favorable to OpenAI would hand every major AI developer — not just OpenAI — a powerful precedent that copyright holders would need years and considerable resources to challenge.

For now, the copyright question remains before Judge Stein, unresolved. The equity question remains a matter of “early conversations,” in the administration’s own description. Whether the two threads are related is something the DOJ’s filing simply doesn’t address — and given the numbers involved, that omission is likely to draw more scrutiny in the days ahead.