Federal Appeals Court Rules AI Company Infringed Copyright by Training on Competitor's Legal Database
The Third Circuit Court of Appeals rejected an artificial intelligence startup's fair use defense in a copyright infringement case brought by Thomson Reuters over the unauthorized use of legal headnotes from its Westlaw database. The court found that ROSS Intelligence's practice of using a third party to access competitor data specifically to create a rival product constituted copyright infringement rather than permissible fair use. The decision potentially establishes significant precedent for how copyright law applies to AI training practices.
Thomson Reuters initiated legal action against ROSS Intelligence in 2020, alleging that the startup had circumvented access restrictions by employing an intermediary service to obtain proprietary headnotes from Westlaw after being denied direct entry. The lower court determined that Westlaw's headnotes—concise summaries of legal principles preceding court decisions—qualified as original creative works due to the editorial judgment involved in their selection and arrangement, comparable to artistic curation of raw material.
The Third Circuit's affirmation in September 2026 emphasized that the case fundamentally involves conventional copyright principles rather than novel questions about artificial intelligence technology. Judge Tamika Montgomery-Reeves, writing for the panel, rejected arguments that the headnotes lacked sufficient originality for copyright protection and found that ROSS's use was not transformative given its direct application toward creating a competing commercial product.
This decision may influence how AI developers approach training datasets involving proprietary commercial information. Companies offering specialized databases or curated content could gain stronger legal protections against unauthorized machine learning applications, potentially affecting the accessibility and cost of AI development. Conversely, the ruling may constrain certain AI innovation practices while creating liability concerns for startups using third-party data sources. The precedent may prompt clearer licensing frameworks between content providers and AI companies, though it could also raise questions about fair use boundaries in different technological contexts.