Third Circuit upholds ruling that Ross's AI training was not fair use

Here's the headline authors have been waiting on: a federal appeals court just upheld a finding that Ross Intelligence's use of Westlaw headnotes to train its AI legal search tool was not "fair use." That's the Third Circuit, ruling on the Ross Intelligence case, and it matters well beyond the parties involved.

For two years, every AI copyright lawsuit — and there are many — has leaned on one unresolved question: does training a model on someone's book count as "transformative" use the law protects, or is it just unauthorised copying with extra steps? Publishers and tech companies have fought that question in courtroom after courtroom, often with opposite results from different judges. This ruling doesn't settle the whole fight: the court distinguished generative AI cases like Bartz and In re OpenAI rather than deciding them, and both rights holders and AI advocates are claiming the ruling as a win.

And yet — here's the twist, and it's worth sitting with. Copyright Lately's read on the same ruling is blunt: "Everyone won the Ross fair use appeal. Except Ross." Ross Intelligence, the defendant, appealed and lost: the court rejected its fair use defense, and the cost of defending the lawsuit had already put it out of business in early 2021. Rights holders and AI advocates are both claiming the ruling as a win, because they are reading different parts of it.

What this means for you: if you've got a claim in, or are weighing whether to file one, against a company that trained a model on your book without a licence, this ruling is now something your lawyers can cite, but it distinguished generative AI cases rather than deciding them, so expect the other side to say it doesn't apply. It's a Third Circuit decision, which carries weight beyond its own jurisdiction — courts elsewhere will read it, cite it, argue with it. Just don't assume it settles your claim.

Here's what to actually do this week. If you're already part of any AI-training litigation or class action — Anthropic, OpenAI, or others — check with your Guild or counsel on whether this ruling gets cited in your case, and ask directly. If you haven't filed anything and think your work was used to train a model, this is the week to find out whether a claims process exists for you; don't wait for a settlement to announce itself. And if you're reading three different accounts of this same ruling and they don't quite agree on what it means for the losing party versus the legal principle — that's not spin, that's just appellate law being appellate law. Keep watching this one. It's the first appeals-court ruling of its kind, and the next cases will show how far it reaches.

Correction, 7 October 2026: An earlier version of this story said the Third Circuit ruling backed the view that training AI on copyrighted books is infringement, said Ross Intelligence brought the case, and said Ross lost on other grounds. Thomson Reuters sued Ross over Westlaw headnotes used to build a competing, non-generative search tool, Ross lost on fair use, and the court distinguished generative AI cases rather than deciding them. We have corrected the story.