Third Circuit Holds Training a Competing AI Legal Tool on Westlaw Headnotes Is Not Fair Use

 By Imaduddin

The U.S. Court of Appeals for the Third Circuit has affirmed that copying Westlaw headnotes to train a competing AI legal-research tool is not fair use. On 29 September 2026, a unanimous panel upheld partial summary judgment for Thomson Reuters against ROSS Intelligence, marking the first federal appellate decision on whether training an AI system on copyrighted material qualifies as fair use.

ROSS Intelligence, a now-defunct startup, built an AI search engine that answered plain-language legal questions by returning passages from judicial opinions. After Westlaw refused it access, ROSS hired a third party to create training materials. The drafters used thousands of Westlaw headnotes — short editorial summaries of legal points written by Thomson Reuters editors — to frame questions. Those materials were then used to train ROSS’s system. Thomson Reuters sued in the District of Delaware for copyright infringement.

In February 2025, Judge Stephanos Bibas (sitting by designation) granted partial summary judgment to Thomson Reuters. He held that 2,243 headnotes were original enough for copyright protection and that ROSS’s use was not fair use. ROSS appealed. The Third Circuit, in an opinion by Judge Tamika Montgomery-Reeves, affirmed.

The court found the headnotes possessed the requisite originality under the low threshold set by Feist. On fair use, it emphasised that ROSS’s use was highly commercial and only minimally transformative. Both services helped users find relevant law; the AI-training step was an intermediate step toward the same ultimate purpose. The court also found market harm, including to a potential licensing market for headnotes as training data. In a notable footnote, the panel distinguished the case from generative-AI disputes, noting that ROSS’s tool “cannot generate original expression” and was trained “for the purpose of creating a commercial substitute for Westlaw.”

The decision does not resolve the broader generative-AI training cases still pending in other circuits (including the OpenAI multidistrict litigation). It does, however, show that courts will treat commercial, non-transformative copying of editorial content to build a direct competitor as ordinary copyright infringement rather than fair use.

For companies developing AI tools in Asia or elsewhere, the practical lesson is clear: using copyrighted editorial or curated content to train a product that competes with the source is high-risk. Licensing or building on public-domain materials remains the safer path, especially where the output is designed to substitute for the original service.

Sources

• Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 25-2153 (3d Cir. Sept. 29, 2026)
• Reuters, “US appeals court upholds Thomson Reuters' landmark win in AI training lawsuit” (29 September 2026)
• Courthouse News Service, “AI training of copyrighted material not fair use: Third Circuit” (30 September 2026)
• Copyright Lately, “Signed, Sealed, Affirmed: Third Circuit Backs Thomson Reuters over ROSS” (29 September 2026)