DOJ Backs OpenAI Against New York Times on AI Copyright

The United States government has formally entered the biggest copyright fight of the AI era — on the side of the machines. In a statement of interest filed September 1 in Manhattan federal court, the Department of Justice urged the judge overseeing The New York Times' lawsuit against OpenAI and Microsoft to hold that training large language models on copyrighted works is fair use — and warned that treating it as infringement "would be problematic — and legally incorrect."

It is, as far as anyone can tell, the first time the federal government has taken a merits position in any of the dozens of copyright suits pending against AI companies. The filing lands in In re OpenAI, Inc. Copyright Infringement Litigation, the consolidated proceeding before Judge Sidney Stein that bundles the Times' December 2023 suit with cases from book authors and other publishers. It is signed by Associate Attorney General Stanley E. Woodward, Jr., Assistant Attorney General Brett Shumate, and DOJ Senior Counsel Michael Weisbuch, invoking a statute — 28 U.S.C. § 517 — that lets the government speak up in any pending case, invited or not.

Word spread fast, in the way these things now do:

SITUATION DETECTED: The US government has sided with OpenAI against the New York Times. The DoJ says training an LLM on copyrighted works does not violate copyright law, and that treating it as infringement would hurt US science, prosperity, and national security.

— MTS (@MTSlive) September 2, 2026

That summary — science, prosperity, national security — matches the headlines. It does not quite match the document. We ran a term-frequency count over the text of the 20-page filing, and the numbers tell a more lawyerly story.

What the brief says versus what people think it says

Term	Occurrences in the filing
"fair use"	31
market	23
transformative	17
substitute / substitution	16
competition / competitive	11
licence / licensing	10
national security	9
innovation	8
science / scientific	5
foreign adversaries	2
journalism / journalistic	1
China	0

Method: case-insensitive stem counts over text extracted from the court PDF of Dkt. 316, a filing of roughly 6,800 words.

"National security" appears nine times in the 20-page filing; "China" appears zero times; "journalism" exactly once. The heart of the brief is not saber-rattling but two technical fair-use arguments. First, that training is "extraordinarily transformative" — the copies exist to teach a model statistical relationships between words, not to republish anyone's articles. Second, that the market-harm question courts must ask is not whether publishers lose some readers, but whether the AI use creates "significant substitutive competition" for the original works — and since ChatGPT does not hand users the Times articles it trained on, the government argues, it does not.

The national-security material is real but compact: quoting a Government Accountability Office report on AI in intelligence analysis and weapons systems, the brief warns that rules hobbling US AI development "threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered." The adversaries go unnamed.

The sharpest passage is aimed at another judge

The most consequential section of the brief may be its attack on Kadrey v. Meta, the Northern California case in which Judge Vince Chhabria sided with Meta on the bottom line last year but mused that AI models could flood markets with machine-made books that "dilute" demand for human ones. Publishers and authors have leaned on that market-dilution language ever since. The DOJ calls it "contrary dicta" that "misapplies copyright principles to LLM training," adopted "without the benefit of briefing," and urges Judge Stein not to follow it. For the music industry's parallel fights — where labels are suing AI song generators on exactly this substitution theory — that passage may matter more than anything the brief says about newspapers, as Billboard's coverage notes.

Notably, the government also hedges in ways the loudest cheerleaders missed. A footnote takes "no position on whether a licensing regime would be financially or logistically feasible," and the brief goes out of its way to bless the voluntary licensing deals publishers have signed for paywalled and real-time content. Training on what you can lawfully read is fair use, in the government's telling; nothing in the filing says acquiring content unlawfully is — a distinction that just cost Anthropic $1.5 billion in the Bartz settlement over pirated book libraries, as TechCrunch points out.

The Times, and the internet, respond

The New York Times came out swinging. 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 spokesperson said in a statement carried by The Hollywood Reporter. "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."

Under the viral post announcing the news, the replies quickly converged on two theories. The most common was geopolitical — "Major win for AI — China doesn't care about copyright law either," ran one of the most-liked responses — which is a reasonable read of the administration's motives, even though, per the count above, the word China never appears in the brief itself. The second theory was darker: that copyright is being de facto abolished only for those with scale. "You, peasant, will be charged with 'theft' of ideas because you are not in the Big Club," one reply put it. The filing gives that camp something to work with — it defends training, not outputs, and leaves every output-side infringement claim alive — but it also directly disputes the "big club" framing, arguing that mandatory licensing is what would entrench giants, since "only the largest technology companies might have the capital necessary to pay licensing fees" while the fees would act "primarily as large subsidies for old mainstream media companies." The rest of the reply section was, as one might expect, mostly cheering for the slop cannon.

Why it matters beyond one lawsuit

A statement of interest binds no one. But federal judges read what the Solicitor General's building sends them, and this one arrives while the central legal question of the generative-AI economy — can you build a frontier model on the open written record without paying for it? — is unresolved in every US appellate court. The brief's citation strategy is tuned for its audience: it anchors on Authors Guild v. Google, the Second Circuit's Google Books ruling that binds Judge Stein, while never once mentioning Thomson Reuters v. Ross, the only US decision so far to reject fair use for AI training. If Judge Stein follows the government's framing, publishers' strongest remaining ground shifts from training to outputs and acquisition. If he rejects it, the United States will have argued for the losing side in the defining copyright case of the decade — the full docket is here, and the next major rulings are expected as the consolidated cases move toward summary judgment.

Photo: lower Manhattan's Civic Center — home of the federal courts where the OpenAI copyright litigation is being heard — by jesse orrico, via Wikimedia Commons (CC0, public domain).