← All stories

Law and policy

The Justice Department backs OpenAI on fair use; weeks later an appeals court rejects fair use in a narrower AI case

On September 1 the Justice Department told a court that training AI on copyrighted work is fair use, surprising two other federal agencies; on September 29 an appeals court rejected fair use in a case about a non-generative legal-research tool.

By · News from Sep 29, 2026 · 4 min read · Checked Oct 2, 2026

AI models learn from enormous amounts of text and images, much of it made by people who were never asked. Is that legal? In September 2026, two very different signals arrived, four weeks apart. They point in different directions, but they are not a straight contradiction, because they concern very different cases.

The big question in plain English

Copyright is the law that gives creators control over copies of their work. Training an AI model means feeding it huge amounts of material so it can learn patterns. To do that, AI companies make copies of articles, books and pictures.

Their main defense is fair use. This is a part of US law that allows people to use copyrighted material without permission in some situations. Judges decide case by case. One thing they look at is whether the new use is transformative, meaning it does something genuinely new with the material rather than simply replacing the original.

September 1: the government picks a side

On September 1, the US Department of Justice filed a 20-page statement of interest in the New York Times's lawsuit against Microsoft and OpenAI in Manhattan federal court. A statement of interest is a document in which the government tells a judge its view of a case it is not directly part of. It was the federal government's first formal position in the AI copyright cases.

The filing argued that training AI is so strongly transformative that it counts as fair use. It went further, saying licensing barriers "function primarily as large subsidies for old mainstream media companies", according to gHacks. It also argued that restrictive rules would threaten national security and help foreign rivals.

The filing did not speak for the whole government. Axios reported on September 19 that it took two agencies that normally weigh in on copyright, the Copyright Office and the US Patent and Trademark Office, by surprise. Both declined to comment. Later in September, the New York Times and other news outlets filed a response arguing that the Justice Department's view clashes with the Copyright Office's own analysis, MLex reported.

One important detail: a statement of interest is non-binding. The judge can weigh it, but does not have to follow it.

September 4: two newspapers ask for AI models to be destroyed

On Friday September 4, the Seattle Times and Newsday sued OpenAI and Microsoft in federal court in New York, Reuters reported. They allege that their articles, including ones behind paywalls, were copied to train ChatGPT, Copilot and Bing. Alongside damages, they asked the court to order the destruction of copies of their work, and of training datasets or AI models that include it.

Microsoft said it was surprised by the lawsuit. OpenAI said its training relies on publicly available data and is protected by fair use.

September 29: an appeals court rules the other way, in a narrower case

An appeals court is a higher court that reviews decisions made by a lower court. Its rulings carry extra weight, because other courts look to them.

On Tuesday September 29, the 3rd US Circuit Court of Appeals upheld Thomson Reuters' win over a company called Ross Intelligence. The written opinion was briefly kept sealed while the parties had a chance to ask for parts to be blacked out; neither did, and it was released on September 30. Thomson Reuters runs Westlaw, a legal research service whose editors write short summaries of legal points called headnotes. Ross used those headnotes to train a competing legal search tool. Thomson Reuters sued in 2020, and Ross shut down in 2021, citing the cost of the case. The appeals court agreed that Ross's use was not transformative, because it was building a direct substitute for Westlaw, and so it was not fair use.

Reuters described it as the first ruling of its kind by a US appeals court in the wave of AI-training copyright cases. Thomson Reuters said that "respecting copyright is essential for fostering innovation".

A crucial caveat: this case did not involve generative AI, the kind that writes text or makes images, like ChatGPT. Ross's tool was a legal-research search tool that competed directly with the service it copied from, and the opinion itself sets this case apart from the generative-AI cases. The ruling does not mean ChatGPT is illegal.

Why the two moves don't cancel out

The Justice Department offered an opinion that the judge is free to set aside, and that other parts of the government did not sign up to. The appeals court made a binding decision, but in a narrower case about a different kind of AI. The big generative-AI cases, including the New York Times's, are still open, and judges will settle them.

What we don't know yet

Sources

Keep reading

Law and policy

Slowing AI down for safety is an illegal cartel, a new lawsuit claims

On September 18 four paying subscribers filed a proposed class action accusing Anthropic, OpenAI, SpaceXAI and Google of breaking US competition law by backing a plan to pace AI development.

Sep 18, 2026 · 4 min read

New to AI? Start with which assistant to use, how to stop AI training on your chats, which ChatGPT plan to pay for or how to write a good prompt.