Heppe-Smith Publishing

← Back to blog

OpenAI authors case: the US government says AI training is fair use. What it means for you

James Heppe-Smith

Stack of leather-bound books beside a legal folder tied with red ribbon on a polished desk, with glowing circuit lines reflected in the wood

On 1 September the US Department of Justice stepped into the OpenAI authors case, the biggest copyright fight authors have with an AI company, and it didn’t take the authors’ side. In a filing in the consolidated case in New York, the government argued that training AI models on books is fair use, calling it “extraordinarily transformative.”

That sounds like a turning point. It isn’t, or not yet. Here is what the filing is and isn’t, where the OpenAI authors case stands, and what it means for you, whether you publish in the US, the UK or both.

What the government filed

The document is a Statement of Interest. A federal law lets the Justice Department tell a court what the United States thinks about a case it isn’t a party to, and that is what this is. It went in under In re OpenAI, Inc. Copyright Infringement Litigation, the case before Judge Sidney Stein in the Southern District of New York that bundles the news publishers’ claims with the Authors Guild’s class action for book authors.

Reports of the 20-page filing put the government’s argument in three parts:

  • Training is transformative. Copying books to build a model serves a different purpose from the books themselves, so it should count as fair use.
  • Training doesn’t replace the books. The government’s view is that training on a work, taken on its own, doesn’t generally produce a substitute for it.
  • Licensing would help only the biggest firms. IPWatchdog reports the filing’s warning that requiring licences “could create entry barriers that only the largest technology companies could absorb,” and that it frames the question partly as one of American competitiveness against China.

One limit matters. The argument covers training, and only training. As dig.watch summarises it, the government leaves open that some uses may not be transformative where a model reproduces and spreads an original copyrighted work. A model that spits out chunks of your book is a separate question.

Why it isn’t a ruling

A Statement of Interest is advocacy. It isn’t evidence, it isn’t law and it doesn’t bind the judge. Judge Stein can adopt its reasoning, ignore it or take the parts he finds useful. What it does tell us is where the current administration stands on AI and copyright, in this case and probably in others to come.

Where the OpenAI authors case stands

The case has reached summary judgment, the stage where each side asks the judge to settle the main questions without a trial. OpenAI filed its motion on 4 September, and the Authors Guild says it and its co-plaintiffs filed theirs the same day. On 24 September the judge moved the remaining deadlines back by two weeks for every party: opposition briefs are now due on 23 October, friend-of-the-court briefs on 30 October and replies on 20 November. No hearing date has been reported.

On 21 September the Authors Guild also released newly unsealed briefs from the case. The Guild says they show OpenAI and Microsoft executives knew about the use of pirated book libraries and went ahead anyway. That is the Guild’s reading of discovery documents, an allegation rather than a finding by the court. It matters because it goes to how the books were obtained, and that is where authors have done best so far.

What the courts have said so far

Two rulings from June 2025 set the scene. Both left a door open.

  • Bartz v. Anthropic. Judge William Alsup found training on lawfully bought books to be fair use, and building a library of millions of pirated books not to be. The piracy finding is what led to Anthropic’s $1.5 billion settlement with authors.
  • Kadrey v. Meta. Judge Vince Chhabria ruled for Meta on training, then said the authors might well have won on an argument they never made: that AI output floods the market and dilutes demand for human-written books.

Added 1 October 2026. An appeal court has since ruled on a narrower question. On 29 September the Third Circuit held in Thomson Reuters v. ROSS Intelligence that copying Westlaw’s headnotes to train a legal search tool was not fair use. The court stressed that ROSS’s tool was not generative AI, and in a footnote it said of the Justice Department’s OpenAI filing that the concerns raised there “do not apply here.” So a copyright owner has won on appeal, and the question in the OpenAI authors case is where it was.

That second door is the one to watch. The government’s claim that training doesn’t produce substitutes reads like an attempt to shut it before Judge Stein gets there. Whether he agrees is the open question.

What it means for you

Nothing changes today. Your copyright is as strong as it was in August. The filing is an opinion offered to a judge who hasn’t decided anything yet.

If you claimed in the Anthropic settlement, that is the one pot of money heading to authors right now. The first distribution is expected in November. The claims deadline passed on 30 March 2026, so there is no joining it now. The Authors Guild’s settlement guide has the detail.

If you publish in the US, registration matters. In American copyright cases, whether you registered a work with the US Copyright Office affects what you can claim if it is infringed. Plenty of indie authors never do. The Copyright Office’s registration pages explain how.

If you’re in the UK, no new law is coming soon. In March 2026 the UK government published its report on copyright and AI and dropped its earlier plan for an exception that would have let AI firms train on works unless creators opted out. It hasn’t legislated and says it will gather more evidence. So the existing law stands, and the US cases matter more to UK authors than they might appear to, because that is where the big AI companies are being tested.

In short

The US government has told the judge in the OpenAI authors case that it thinks AI training on books is fair use. That is a strong signal, not a decision. The case is mid-briefing, the question of AI output crowding out books is still open, and the courts have already drawn a hard line on piracy. Keep your rights paperwork in order and watch for Judge Stein’s ruling. That is the one that counts.

For more on protecting your author business, see our pieces on the scam emails that quote your own book and this year’s KDP and ACX changes.

Sources

← More posts