Appeals court hands Thomson Reuters first federal AI copyright win
The 3rd Circuit rejected Ross Intelligence's fair use defense but said its reasoning may not stretch to generative AI tools like Claude or ChatGPT
Here's the headline risk: this sounds like a win for every author suing an AI company, and it isn't quite that. Judge Montgomery-Reeves drew a sharp line — Ross's tool retrieved and competed, it didn't generate original text, and that's exactly why it lost. Generative AI makers now have a roadmap for arguing their tools sit on the other side of that line. Read the ruling closely before you cite it as precedent for your own case.
A federal appeals court just handed publishers their first real win against an AI company — and then spent several paragraphs warning you not to get too excited about it.
On Tuesday, the 3rd U.S. Circuit Court of Appeals upheld Thomson Reuters' 2025 victory over Ross Intelligence, a now-defunct legal AI startup that trained its search engine on thousands of Westlaw "headnotes" — the short summaries Thomson Reuters' editors write to explain legal issues in court opinions. Judge Tamika Montgomery-Reeves wrote that the material carried "a creative spark" and that Ross "aspired to be a direct competitor by using them for a highly similar purpose." No fair use defense survives that combination: copied creative work, built into a rival product.
This is the first federal appeals case anywhere to reject a fair use argument for training a commercial AI product on publisher content without permission or payment. Thomson Reuters filed the suit back in 2020, before ChatGPT existed, and judge Stephanos Bibas ruled in the company's favor in February 2025.
Here's the catch, and it's the whole story: Montgomery-Reeves went out of her way to say this case doesn't concern large language models. Ross's tool retrieved existing passages — it didn't generate original text. She explicitly contrasted that with Bartz v. Anthropic, the case where a judge ruled last year that training Claude on published books was fair use, because Anthropic's model "cannot" be compared to a search tool that just surfaces someone else's sentences. Anthropic lost that case only on the piracy question — the pirated copies, not the training itself — and that loss produced the $1.5 billion settlement authors are still collecting from.
So what does this mean for you if you write books, not legal headnotes? It means the legal ground under generative AI training is still largely fair-use-friendly, according to the one ruling we have on the books. This Thomson Reuters win is powerful precedent if an AI company builds something that competes directly with your work using your exact material — think a tool that just serves up your paragraphs. It is not, on its own, precedent against the Claudes and ChatGPTs of the world, and the judge said so in writing.
What to do this week: if you're part of any ongoing or future AI litigation, don't assume this ruling carries your case. Ask your Guild rep or attorney specifically whether your claim resembles Ross — direct substitution, same expression — or resembles Anthropic — transformative generation. That distinction is now the line a federal appeals court has drawn, and it will decide which side of future cases you're standing on.
Why it matters to you: The first federal appeals ruling on AI and fair use favors publishers whose work was copied to build a direct competitor, but the judge says it may not apply to generative AI models at all.
This story ran in the Coral Wire edition of Sat, Oct 3 2026, when the Coral Index stood at 62/100 (Tailwinds building).
Coverage — 6 outlets
Every outlet found reporting this story.
- What does Thomson Reuters' landmark AI copyright win tell us about fair use and training data?legallayer.substack.com
- Federal Court Just Dealt a Massive Blow to Publishers Suing Google AIandroidheadlines.com
- Federal Court Dismisses Publishers’ Suits Over Google’s AI Overviewstheinformation.com
- Google AI Overviews Lawsuit: Publishers Lose US Caselapaasvoice.com
- AAP, Authors Guild Applaud Decision in ROSS Copyright Infringement Casepublishersweekly.com
- Federal appeals court upholds Thomson Reuters’ landmark AI copyright win | Nieman Journalism Labniemanlab.org
Also in this edition
- Amazon slashes KDP upload limit from ten titles to twoPlatform — The cap, effective Monday September 21, applies per format per week with no exceptions for reverted rights or backlists
- Two authors sue LA library over canceled Palestinian eventLegal — A petition drew 7,000 letters and 100-plus LAPL staff signed a protest letter before Matari and Murad filed suit
- AI company settles copyright suit against MicrosoftLegal — The case ends in a deal rather than a ruling, so the legal questions at its core go untested
- Tariffs and plant closure push books past pub dateMarket — Josh Silver's Fruit Fly and Marlon James's The Disappearers are among the titles caught in a fall supply squeeze.
- InkShares founder explains crowdfund-to-Hollywood pipelineMarket — Adam Gomolin points to The God in the Shed topping 100,000 units as proof the preorder model can find writers traditional houses pass on
- Massachusetts signs book-ban firewall after year-long author pushLegal — The law leans on Rhode Island's year-old S.0238 as its blueprint and limits challenges to school staff, caregivers, and students
- Sold becomes the most challenged book amid bansLegal — Author Patricia McCormick's publisher has built a revised educator guide and 20th-anniversary video to keep the title in classrooms.
- Audible adds AI chat features starting with Dracula spinoffPlatform — The beta starts with Audible Originals, letting listeners talk to a generative-AI version of Renfield from a Dracula adaptation this fall.