Appeals court rules Thomson Reuters headnotes beat fair use defense
The Third Circuit's Oct. 7 ruling is the first federal appellate decision to address fair use in an AI-training case, and Ross lost
Don't let the breathless takes fool you — this one is narrow. A panel of judges looked at Thomson Reuters' Westlaw headnotes, found they had enough original content to be copyrighted, and said Ross's non-generative tool training on them wasn't fair use. That's it. That's the ruling. It's still the first time an appeals court has said anything at all on this question, which is exactly why every lawyer on every side of the 100-plus pending AI copyright suits is going to be reading it line by line this week.
Here's a ruling worth reading slowly, because the headline writers are already overselling it.
On Wednesday, Oct. 7, the Third Circuit sided with Thomson Reuters against Ross Intelligence, upholding a lower court finding that Reuters' Westlaw headnotes — those short summaries of legal issues that sit above judicial opinions — have enough original content to be copyrighted. Ross had used a third-party tool to scrape that content off Westlaw and train its own competing legal research tool. The court said no, that's not fair use. Circuit Judge Tamika Montgomery-Reeves put it plainly for the panel: "In truth, this is no more than an ordinary copyright case."
That line matters as much as the outcome. The court went out of its way to say it wasn't ruling on "the future of AI legal technology" — it was ruling on a specific set of headnotes and a specific, non-generative tool. Ross doesn't even exist anymore; defending this lawsuit put it out of business back in 2021. According to the Copyright Lately blog, everyone involved is claiming this as a win except Ross — whose lawyer, Yar Chaikovsky of White & Case LLP, told Courthouse News they intend to seek Supreme Court review "to obtain much-needed clarity on these issues."
What this means for you. If you're an author or publisher watching the AI copyright fights from the sidelines, this is the first time a federal appeals court has weighed in on fair use in an AI-training dispute at all. It's not a sweeping win for creators generally — it's a win for Thomson Reuters, on Thomson Reuters' specific facts. But it is now a real citation sitting in the record, in a landscape where more than 100 copyright-infringement lawsuits are active against AI companies, spanning music, newspapers, comedians, and more. One of those, Concord Music Group v. Anthropic, over Claude's training on copyrighted lyrics, is expected to be heard this month in the Northern District of California.
Worth remembering too: this summer's $1.5 billion Anthropic settlement came out of a case where the court found training language models on books can be fair use — but ruled that storing millions of pirated books in a central library was not. Different facts, different outcome. That's the pattern across all of this right now: judges ruling narrowly, on the specifics in front of them, not handing down one grand verdict on AI training.
What to do this week. If your work is part of any pending AI-training litigation, read the Ross ruling's actual language on originality and headnotes — not the summary. The specifics of what made Reuters' content copyrightable are exactly the kind of detail your own case may hinge on.
Why it matters to you: This is the first federal appellate ruling to address fair use in the AI-training context, so it's now a data point every pending copyright case against an AI company will have to reckon with.
This story ran in the Coral Wire edition of Sun, Oct 11 2026, when the Coral Index stood at 80/100 (Green light).
Coverage — 1 source
Every outlet found reporting this story.
Also in this edition
- Smashwords adds pre-orders as Spotify audiobooks hit 180 countriesPlatform — Smashwords pre-orders can run up to twelve months out, while Spotify plans to take audiobooks from 22 countries to the full global market by year's end.
- Three of five big publishers allegedly use AI without telling authorsMarket — Wired's anonymous sourcing says the undisclosed AI use spans cover copy, PR text, and cover art itself
- USA Today publisher sues OpenAI for over $250 millionLegal — The suit covers 19 publications and alleges OpenAI's internal communications show it knew its models could reproduce news content
- Goodreads adds half and quarter stars after a decade of askingPlatform — The site added 3.5 million new ratings in under two days, and the top new score so far is 3.5