A US appeals court has handed copyright owners their first appellate win over AI training. On 29 September 2026, the Third Circuit affirmed the lower court's judgment for Thomson Reuters against Ross Intelligence, as Copyright Lately reports. Ross can't rely on fair use for the Westlaw material it used to train its legal research tool.
Key takeaways
The Third Circuit upheld the finding that Ross Intelligence infringed Thomson Reuters' copyright in Westlaw headnotes.
It's the first federal appellate ruling on fair use in AI training.
The case involves non-generative AI, and the full opinion is still sealed, so its reach is not yet clear.
What is fair use?
Fair use is a US legal doctrine that allows limited use of copyrighted material without permission, depending on factors such as the purpose of the use and the harm to the original's market. Most AI developers lean on it to justify training on copyrighted content.
The case in brief
Ross was a legal research startup, now defunct, that built an AI tool using material derived from Westlaw headnotes. Headnotes are the short summaries Thomson Reuters' editors write for court opinions.
In February 2025, trial judge Stephanos Bibas ruled against Ross, finding it had infringed 2,243 headnotes. The appeal went to a three-judge panel, and Judge Tamika Montgomery-Reeves wrote the decision affirming that result.
Why the court sided with Thomson Reuters
The appellate opinion remains sealed, so the detailed reasoning isn't public. The trial ruling it affirmed rested on three points:
Headnotes are original expression and can be protected, even when they draw on public court opinions.
Ross's use wasn't transformative, because it built a direct competitor to Westlaw.
The harm to Thomson Reuters' market weighed against fair use.
The limits of the ruling
This is a narrower precedent than the headline suggests. Ross's tool wasn't generative AI. It retrieved existing judicial opinions and didn't write new text. Copyright Lately notes that several questions stay open until the opinion is unsealed, including whether a potential licensing market is enough to show harm.
Even so, the ruling will be cited in the dozens of generative AI cases now pending in US courts.
Why it matters
Any business that trains or fine-tunes models on third-party content should take note. The direction is towards licensing, provenance and clear records of where data came from. That applies just as much to companies in Malaysia and Singapore building products for the US market.
It also lands in the same week Washington chose voluntary self-regulation for AI. Courts, not pledges, are setting the hard rules. For firms moving generative AI from pilot to production, data rights belong on the checklist next to accuracy and cost.
Frequently asked questions
What did the Third Circuit decide?
It affirmed the judgment for Thomson Reuters, agreeing that Ross Intelligence infringed copyright and couldn't claim fair use.
Does the ruling apply to generative AI such as chatbots?
Not directly. The case involved a non-generative legal search tool. It will still influence pending generative AI cases.
What are Westlaw headnotes?
Short editorial summaries of points of law in court opinions, written by Thomson Reuters' editors.
What happened to Ross Intelligence?
The startup is no longer operating.