Third Circuit Rejects Fair Use for AI Training on Westlaw

Copyright disputes involving AI training have historically turned on the interpretation of 'transformative use' under the fair use doctrine. Prior district court decisions have produced divergent outcomes, with some courts finding that machine learning extraction serves a different purpose than human reading, while others emphasize the competitive harm to licensing markets. In the legal information sector specifically, headnotes and case summaries have long been treated as protectable creative works due to the editorial judgment involved in their creation, distinguishing them from raw government opinions which lack copyright protection. Appellate courts have previously addressed similar issues in non-AI contexts, such as database scraping and search engine indexing, often focusing on whether the secondary use supplants the primary market. The Third Circuit has traditionally applied a rigorous standard for transformative use, requiring clear evidence that the new work adds distinct expression or meaning rather than merely repackaging existing content. This pattern suggests that industries relying on curated, proprietary datasets face higher litigation risk when those datasets are used without permission for model training, particularly when the training directly enables competing commercial products.

On October 7, 2026, the U.S. Court of Appeals for the Third Circuit issued a binding appellate opinion affirming copyright protection for Westlaw headnotes and explicitly rejecting the fair use defense in the context of artificial intelligence training. According to multiple legal-focused Bluesky accounts that disseminated the ruling simultaneously on the day of issuance, the court upheld a lower court decision in favor of Thomson Reuters, establishing that the unauthorized ingestion of proprietary legal data for AI model development constitutes actionable copyright infringement under current Third Circuit jurisprudence.

The dispute centered on whether an unnamed AI defendant’s use of Westlaw’s proprietary headnotes qualified as fair use under U.S. copyright law. The defendant asserted that their AI training process was transformative, arguing that the system did not merely reproduce the headnotes but instead converted them into new functional capabilities distinct from the original expressive content. This argument relied on the premise that machine learning models extract statistical patterns and semantic relationships rather than republishing creative expression, thereby serving a fundamentally different purpose than the original human-authored legal summaries.

Thomson Reuters countered this position by arguing that the unauthorized copying of headnotes during the training phase itself constituted infringement, regardless of the downstream output's functionality. According to the sources reporting on the decision, the Third Circuit sided with Thomson Reuters, determining that the AI training use was not sufficiently transformative to qualify for fair use protection. The court’s reasoning, as summarized in the available reports, emphasized that the headnotes themselves are protected creative works and that their wholesale ingestion for commercial AI development competes with the market for licensed legal data products.

The timing of the decision’s dissemination suggests immediate industry attention. Legal news outlets and specialized intellectual property accounts reported the appellate decision across platforms within seconds of its release at 13:50 UTC on October 7, 2026. The simultaneous posting by three distinct legal-focused Bluesky profiles—ai-law-news.bsky.social, techlawjdsupra.bsky.social, and ip-law.bsky.social—indicates coordinated monitoring of federal appellate dockets by legal technology observers. All three sources utilized identical phrasing to describe the holding: “Third Circuit Affirms Copyright Protection for Westlaw Headnotes and Rejects Fair Use Defense in AI Training Case.”

This ruling represents a significant judicial clarification regarding the intersection of generative AI and copyright law in the legal information sector. While the specific identity of the AI defendant remains unnamed in the provided source materials, the precedent applies broadly to any entity ingesting proprietary legal databases without authorization within the Third Circuit’s jurisdiction. The decision distinguishes between the functional outputs of AI systems and the copyrighted inputs required to train them, suggesting that courts may scrutinize the training process independently from the end-user experience.

Legal observers note that this appellate affirmation creates binding precedent within Pennsylvania, New Jersey, Delaware, and the Virgin Islands. The rejection of the transformative use argument specifically for legal headnotes signals that courts may be less receptive to fair use defenses when the underlying material is highly structured, professionally authored, and commercially licensed. The ruling does not address whether other types of training data or different factual contexts might yield different fair use outcomes, leaving open questions about the decision’s applicability beyond the legal research domain.

The available sources confirm the procedural outcome and the core legal holding but do not provide the full text of the opinion, the specific legal tests applied, or detailed quotations from the judges. Consequently, while the rejection of fair use is confirmed, the precise doctrinal reasoning—such as how the court weighed the four fair use factors or distinguished prior circuit precedents—remains outside the scope of the currently available evidence ledger. Further analysis would require access to the complete slip opinion to determine whether the ruling establishes a categorical bar against AI training on proprietary legal data or adopts a more fact-specific balancing approach.

AI companies developing legal research tools within the Third Circuit must now license Westlaw headnotes or risk infringement liability, increasing operational costs and potentially limiting training data diversity. Thomson Reuters secures strengthened negotiating position for data licensing agreements and can pursue damages against unlicensed users. Legal professionals may face higher subscription fees as AI vendors pass through licensing costs. Competitors using alternative or synthetic training data gain relative advantage. The ruling's geographic scope covers Pennsylvania, New Jersey, Delaware, and the Virgin Islands, creating regional compliance asymmetry until other circuits rule similarly or diverge.