Facts
Thomson Reuters owns Westlaw, whose editorial headnotes summarize points of law and are organized under the Key Number System. ROSS Intelligence, a competitor building an AI-powered legal search engine, asked to license Westlaw content and was refused. ROSS then obtained training data from LegalEase in the form of “Bulk Memos”, legal questions with good and bad answers written by lawyers who worked from a guide based on Westlaw headnotes. ROSS used the memos to train its tool, which returns existing judicial opinions in response to a user’s question. Thomson Reuters sued ROSS for copyright infringement in the U.S. District Court for the District of Delaware.
Question
After a 2023 ruling that left most questions for a jury, Judge Stephanos Bibas, a Third Circuit judge sitting by designation, revisited the cross-motions for summary judgment. He first had to decide whether the headnotes were original enough to be protected and whether ROSS had actually copied them. The decisive question was whether using that material to train an AI system qualified as fair use under Section 107 of the Copyright Act, weighing the purpose and character of the use, the nature of the work, the amount used and the effect on the market.
Decision
The court granted partial summary judgment to Thomson Reuters, finding actual copying of 2,243 headnotes, and rejected ROSS’s fair-use defense. On the first factor, the use was commercial and not transformative, because ROSS used the headnotes to build a product that competes with Westlaw and serves the same purpose. The court stressed that ROSS’s tool is not generative AI and distinguished the intermediate-copying cases, where copying was needed to reach unprotected functional elements. The second and third factors favored ROSS. The fourth and most important factor favored Thomson Reuters, since ROSS meant to create a market substitute and the effect on a potential market for AI training data was enough.
Why it matters
This was one of the first U.S. judgments to decide fair use in a dispute over training an AI system on copyrighted material, and it went against the AI developer. The court confined its reasoning to a non-generative tool built by a direct competitor, which leaves open how far it applies to large language models. Judge Bibas certified the fair-use and originality questions for interlocutory appeal, so the Third Circuit is set to give the first appellate ruling on the issue.
Related stages
The Third Circuit accepted the interlocutory appeal (No. 25-2153) and heard oral argument on June 11, 2026 before Judges Restrepo, Montgomery-Reeves and Bove. As of September 2026 no decision had been issued; a ruling is expected in late 2026. For generative models, see Bartz v. Anthropic and Kadrey v. Meta (N.D. Cal., June 2025).