Appeals court rules ROSS copying of Westlaw was not fair use
It is the first time an appeals court has rejected an AI company's fair use defense for unauthorized copying of this kind

Stop calling it theft "disruption." A federal appeals court just said the quiet part in writing: copying a rightsholder's work to train a competing AI product isn't fair use, it's a shortcut around a market that already exists. The Third Circuit even noted that licensing market is "rapidly developing" — which is lawyer-speak for "there was a door, and ROSS climbed through the window instead." This one's a tailwind. Read it, cite it, and remember the name.
Here's a sentence you don't get to write often in this business: an appeals court looked at an AI company's fair use defense and said, flatly, no.
The case is Thomson Reuters v. ROSS, and it's been grinding through the courts since May 2020, when Thomson Reuters sued after ROSS used its Westlaw headnotes to train an AI legal research platform. The district court in Delaware sided with Thomson Reuters. ROSS appealed. The Third Circuit just affirmed — and in doing so, became the first appeals court to rule that this kind of unauthorized copying by an AI company is not protected by fair use.
That "first" matters more than any single headnote dispute. Fair use arguments from AI companies have leaned hard on the idea that training data markets don't really exist yet — that nobody's losing money because nobody was ever going to get paid for this in the first place. The Third Circuit didn't buy it. The court noted that the licensing market for AI training is, in its words, "rapidly developing." That's not a stray observation — it's the hinge the whole ruling turns on, and it's the line AAP CEO Maria Pallante called out directly, saying the decision "supports the premise that a robust licensing market among developers and rightsholders is essential to achieving sustainable and ethical progress in AI technology."
Authors Guild CEO Mary Rasenberger put it in plainer terms: the ruling "vindicates copyright law" and "clarifies that one cannot steal someone else's works to train an AI system that displaces the very works the AI was trained on." She's right that this cuts at something specific — AI companies "try to minimize or flatly deny" that a licensing market exists in their fair use arguments, and now there's appellate precedent saying otherwise.
What this means for you: if you've been wondering whether your backlist, your training-data complaints, or your publisher's licensing negotiations have any legal teeth, this is the teeth. It's not your case and it's not your payout — but it's the precedent your agent, your publisher, and your Guild will be citing the next time an AI company claims your work was fair game.
This week, do two things. First, if you've gotten any licensing inquiry or notice about AI training use of your work, don't let it sit in your inbox — this ruling just made those conversations worth having. Second, watch for the Authors Guild's next statement on how this precedent applies beyond the legal-research niche ROSS operated in. The language here — "rapidly developing" licensing market — is about to show up in a lot of other briefs. Make sure it shows up in arguments made on your behalf, too.
Why it matters to you: The ruling gives authors and publishers their strongest legal precedent yet for arguing that AI training on copyrighted work requires a license, not a shrug.
This story ran in the Coral Wire edition of Sun, Oct 4 2026, when the Coral Index stood at 58/100 (Tailwinds building).
Coverage — 2 sources
Every outlet found reporting this story.
- AAP, Authors Guild Applaud Decision in ROSS Copyright Infringement Casepublishersweekly.com
- What does Thomson Reuters' landmark AI copyright win tell us about fair use and training data?legallayer.substack.com
Also in this edition
- Judge dismisses Chegg and Penske suits against Google AI OverviewsLegal — Five outlets confirm the dismissal; the ruling clears Google's AI search summaries of antitrust claims for now
- Audible adds AI feature letting listeners talk to book charactersAI — Lithub calls the feature creepy, which tells you the reaction is as much a story as the launch itself
- Written Word Media's 2026 survey ties book count to incomeCommunity — Last year's data showed 40% of authors with 25+ titles earned over $5,000 a month, and the new survey is open now