title: "Third Circuit rejects fair-use defense for AI legal training" slug: "third-circuit-rejects-fair-use-defense-for-ai-legal-training" published: "2026-10-21" beat: "Policy" tags: ["Policy"] creator: "Agentry Newsroom" editor: "Susanne Sperling, Editor — Human in the Loop" tools: ["Claude (Anthropic)", "Perplexity Sonar"] creativeWorkStatus: "verified" dateReviewed: "2026-10-21" aiActArticle50: "compliant" humanView: "https://agentry.news/policy/third-circuit-rejects-fair-use-defense-for-ai-legal-training" agentView: "https://agentry.news/agent/third-circuit-rejects-fair-use-defense-for-ai-legal-training"
The U.S. Court of Appeals for the Third Circuit affirmed on September 29, 2026, that Thomson Reuters's Westlaw headnotes are protected by copyright and that ROSS Intelligence's use of them to train an
Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. AI policy.
The U.S. Court of Appeals for the Third Circuit ruled on September 29, 2026, that Thomson Reuters Enterprise Centre GmbH and West Publishing Corporation won their copyright dispute against ROSS Intelligence Inc. over the unauthorized use of 2,243 Westlaw headnotes—editorial summaries of judicial opinions—in training an AI-powered legal-research platform Reuters.
The appellate panel affirmed the partial-summary-judgment ruling from the U.S. District Court for the District of Delaware, holding that Thomson Reuters's headnotes contained sufficient original expression to qualify for copyright protection. ROSS had argued that its use of the headnotes qualified as fair use because the underlying judicial opinions themselves are public domain. The court rejected that defense Stanford Law School Fair Use.
In its analysis, the Third Circuit held that ROSS used the headnotes commercially to develop a direct competitor to Thomson Reuters's legal-research product, that the use was not sufficiently transformative, and that copying the headnotes was unnecessary because the underlying opinions were already publicly available. The court stated: "Thomson Reuters' materials possess a creative spark, and ROSS aspired to be a direct competitor by using them for a highly similar purpose." Reuters
The decision marks a significant boundary-setting moment for AI training data. Rather than treating all machine-learning uses as transformative, the court emphasized that copying curated editorial work to build a substitute product—even when the underlying raw material is public—does not escape copyright law. The reasoning suggests that AI developers cannot rely on fair-use doctrine simply because they are training on published works LawNext.
The ruling affects how AI agents trained on commercial databases and curated content libraries must navigate licensing and data-access agreements. Legal AI platforms, customer-service agents drawing on proprietary knowledge bases, and other agent systems that rely on processed or annotated data face clearer constraints: creating value-added summaries, classifications, or editorial annotations likely qualifies those inputs as protectable works, distinct from the raw facts or raw text beneath them.
The case does not establish a dollar-damage award or penalty in the appellate ruling itself; the available reports do not identify a separate civil penalty or regulatory action FindLaw. The affirmed partial-summary-judgment disposition means the case may proceed to trial on remaining issues, or parties may settle.