The Justice Department tells a court that training LLMs on copyrighted text is fair use, the first time Washington has taken a side

On September 1, 2026, the Department of Justice filed a statement of interest in the consolidated copyright litigation against OpenAI and Microsoft, urging the judge to hold that training large language models on copyrighted text is fair use. It is the first time the federal government has formally staked out a position in the wave of copyright suits brought by publishers, authors, and news organizations against AI companies. The document does not decide anything, but it tells the court, and the industry, where the administration wants the line drawn.

What the filing actually says

The case is In re OpenAI, Inc. Copyright Infringement Litigation, MDL No. 25-md-3143 in the Southern District of New York, the consolidated proceeding led by the New York Times. The government’s argument is that the fair use doctrine already covers the ingestion of copyrighted works to train a model, and that reading copyright law to forbid it would be a mistake with national stakes. “The United States has a strong interest in this Court rejecting any argument that training LLMs on copyrighted texts violates copyright law,” the filing states. It frames the stakes in economic and competitive terms: constraining model development “under a misunderstanding of fair use doctrine would thwart such creative and scientific progress while hindering American prosperity,” and the country has an interest in a “robust and competitive artificial intelligence industry that sets the standard for the practice and procedure of AI use globally.”

The filing was signed by senior DOJ officials, including Associate Attorney General Stanley Woodward Jr. and Assistant Attorney General Brett Shumate, which signals that the position is the department’s, not a single office’s.

Why a statement of interest is not a ruling

The important qualifier is the document type. A statement of interest lets a federal agency tell a court where the government stands without becoming a party to the case, and it carries no binding authority. The judge is free to disregard it. The New York Times reacted sharply to the filing, and the plaintiffs’ core claim, that OpenAI copied their work without a license and that its models can reproduce protected text, is unchanged by the government weighing in.

So the accurate read is narrow: the executive branch has told the court it wants the training-is-fair-use theory to win, and that view now sits in the record of the most closely watched copyright case in the industry. That is influence, not a verdict. It matters because fair use is decided case by case on a four-factor test, and a judge looking for how to weigh the “purpose and character” of model training now has the government’s thumb visibly on one side of the scale.

What it means for developers

The unresolved question underneath every model you build on is whether the training data was lawfully used, and that question does not stop at OpenAI. The same fair use theory reaches every lab that trained on large text corpora, from OpenAI’s GPT to Anthropic’s Claude, Google’s Gemini, and Meta’s Llama. If courts converge on the government’s position, the legal risk that has been priced into frontier models, and the pressure to license training data at scale, eases, which tends to keep model access broad and costs lower. If courts reject it, the economics of building and serving these models shift, and licensing overhead can flow downstream into pricing and availability.

For teams shipping on top of these APIs, the practical effect is not immediate. No ruling changes your data pipeline today. What changes is the risk picture a technical decision-maker should hold: the fair use question is now a live policy fight as much as a legal one, and the outcome will shape whether the current generation of models keeps its current cost structure or absorbs a licensing tax. Procurement and legal teams evaluating a model vendor have a concrete new reason to ask how that vendor is positioned if the fair use theory loses.

What to watch

The near-term signal is the judge’s ruling on the pending motions, where the statement of interest will either be cited or set aside, and whether other plaintiffs in the parallel suits against other labs draw a matching government filing or a contrary one. The deeper signal is whether fair use for model training settles as a doctrine or fractures across circuits, which is the difference between a stable legal foundation for the industry and years of uncertainty. Stackmaven will revisit on or around December 3.

Sources cited

1. United States' Statement of Interest, In re OpenAI, Inc. Copyright Infringement Litigation, MDL No. 25-md-3143 (S.D.N.Y., filed Sept. 1, 2026) (CourtListener/RECAP) storage.courtlistener.com
2. DOJ urges judge to rule for OpenAI, Microsoft in N.Y. Times lawsuit (The Washington Post) www.washingtonpost.com
3. Trump administration backs OpenAI in NY Times copyright suit (Bloomberg Law) news.bloomberglaw.com
4. US government sides with OpenAI on issue of training LLMs on copyrighted material (TechCrunch) techcrunch.com