For the first time, the U.S. government has directly weighed in on whether training AI on copyrighted text is lawful — and its answer is yes. The Department of Justice filed a 20-page Statement of Interest on September 2nd in Manhattan federal court, backing OpenAI against The New York Times. Every AI tool you use daily — ChatGPT, Claude, Gemini — was built on training data that sits at the center of this fight.
A Statement of Interest isn’t a ruling. Think of it as the executive branch dropping a note to the judge saying, “here’s how we read the law.” It doesn’t decide the case, but it signals policy and can move how Judge Sidney Stein — and courts above him — interpret fair use. The DOJ’s position draws on precedent from cases like Google Books, where large-scale scanning for search indexing was deemed transformative rather than infringing, placing this argument within established legal territory rather than uncharted waters.
What This Means for the AI Tools You Actually Use
The outcome of this case could reshape the models you rely on — or the journalism you depend on to understand the world.
The government’s core arguments, drawn directly from the brief, break down to this:
- LLM training is “non-consumptive” — models learn statistical patterns from text rather than storing or republishing articles.
- Training is “extraordinarily transformative“ because outputs are generated language, not copied paragraphs.
- Restricting this, the DOJ argues, would “thwart creative and scientific progress while hindering American prosperity and economic mobility.”
- AI leadership is framed explicitly as a national security interest.
- The brief also notes that many major publishers, like the Times itself, are reported to be experimenting with LLMs in their own workflows — a pointed observation.
“Constraining LLM development under a misunderstanding of fair use doctrine would thwart such creative and scientific progress while hindering American prosperity and economic mobility.” — DOJ Statement of Interest
The Times disagrees. Filed in December 2023 and seeking billions in damages, their lawsuit argues there is nothing transformative about copying millions of articles to build a product that competes directly with their journalism. When ChatGPT outputs closely mimic Times reporting — a claim OpenAI disputes — that reads less like learning from sources and more like substitution. That distinction is where this case gets genuinely hard to call.
A clarifying frame comes from the Anthropic case, where Judge William Alsup compared LLM training to “any reader aspiring to be a writer” — learning from books to create something new, not to replicate them. Anthropic paid a $1.5 billion settlement, but for acquiring books through illegal shadow libraries, not for the training process itself. How you obtained the content turns out to be a legally separate question from what you did with it afterward.
If Judge Stein sides with the DOJ, AI companies gain significant breathing room — no licensing scramble, no cost surge, models stay broadly capable. Many legal and industry analysts say that if the court rejects that view, expect licensed-only datasets, slower iteration, and potentially data monopolies where a handful of platforms lock up the most valuable content. Both outcomes reshape the internet you use every day.
The law hasn’t answered the fundamental question yet: is an AI reading your work to learn genuinely different from an AI copying your work to profit? The government has placed its bet. Every prompt you type is riding on the answer.
