Thomson Reuters v. Ross

The claims

Copyright infringement of Westlaw headnotes; fair use is the central question on appeal

What has happened

ROSS Intelligence built a natural-language legal search engine. Denied a Westlaw license, it bought training material from LegalEase: “Bulk Memos” written from Westlaw headnotes. Thomson Reuters sued in May 2020. ROSS shut down in early 2021, citing the cost of the litigation, but kept defending the case. In September 2023 Judge Bibas largely denied summary judgment and pointed fair use toward trial. He then reversed course. On February 11, 2025 he granted partial summary judgment for Thomson Reuters: direct infringement of 2,243 headnotes, and no fair use as a matter of law. Factors one and four favored Thomson Reuters. The opinion stressed that ROSS’s tool was non-generative and competed with Westlaw, and that a market for AI training data could be harmed. The Third Circuit accepted an interlocutory appeal in mid-2025, with trial deferred. At argument on June 11, 2026, ROSS leaned on the Third Circuit’s recent fair-use decision in ASTM v. UpCodes; the panel pressed both sides on transformativeness and market harm.

Key developments

• 2020-05-06 — Thomson Reuters sues ROSS in the District of Delaware over training data derived from Westlaw headnotes.
• 2023-09-25 — Judge Bibas largely denies summary judgment and leaves fair use for a jury.
• 2025-02-11 — Revised summary judgment: direct infringement of 2,243 headnotes; fair use rejected, with factors one and four favoring Thomson Reuters.
• 2025-06 — Third Circuit accepts ROSS’s interlocutory appeal (No. 25-2153); trial is deferred while the appeal proceeds.
• 2025-10 — Amicus briefs supporting ROSS’s fair-use position are filed at the Third Circuit.
• 2026-06-11 — Oral argument before Judges Restrepo, Montgomery-Reeves and Bove. Questioning centers on factors one and four; ROSS invokes ASTM v. UpCodes. Decision pending.

Why it matters for training data

This is the first federal appellate test of fair use for AI training, and the only training case so far where a court has rejected fair use outright. If the Third Circuit affirms, buying compiled training data derived from a rival’s copyrighted product is high-risk even when outputs reproduce none of the original text, and the “market for training data” becomes a recognized factor-four harm that strengthens every licensor’s pricing position. The district court expressly limited its reasoning to non-generative AI, so how much of the ruling carries over to generative models is the open question buyers and suppliers should watch.

Sources

• LawSites: Third Circuit oral argument report (June 2026)
• Third Circuit docket, No. 25-2153 (CourtListener)
• Authors Alliance analysis of the February 2025 fair-use ruling
• Baker Botts: Third Circuit hears oral argument (July 2026)

Deeper analysis

• Thomson Reuters v. Ross, explained
• Training-data copyright lawsuits: the landscape

← All case files Weekly lawsuit news