Appeals Court Upholds Thomson Reuters’ Copyright Win Over AI Training

Using copyrighted legal summaries to train a competing AI search tool did not qualify as fair use, the Third Circuit has ruled. The court upheld Thomson Reuters’ victory against Ross Intelligence on September 29, 2026. Its opinion became public the next day, explaining a decision that marks the first federal appellate ruling directly addressing fair use in AI training.

The opinion, detailed in an October 5 analysis by Jackson Walker attorney Emilio B. Nicolas, replaces the initial picture of a verdict with sealed reasoning. The judges found that Westlaw’s headnotes qualified for copyright protection and that three of the four statutory fair-use factors weighed against Ross. Fair use is a defense that can excuse an otherwise infringing use of copyrighted material.

A new tool, but the same research purpose

The dispute concerned more than 2,000 Westlaw headnotes: summaries of legal points and court holdings used to help researchers find relevant cases. Ross used that material to train its own legal-research system. On appeal, it challenged both the headnotes’ eligibility for copyright protection and the rejection of its fair-use defense.

The court rejected both arguments. It found enough creativity in the headnotes to protect them and described Ross’s use as minimally transformative at best. A transformative use adds a different purpose or character rather than merely replacing the original’s function. Here, the judges focused on what the two products ultimately did, not simply on Ross’s use of AI.

to create and optimize a legal-research platform that helps users find responsive legal material.

Third Circuit, describing the shared purpose of the companies’ uses, as quoted by Jackson Walker

The boundary around generative AI

Ross’s system returned passages from existing judicial opinions in response to legal questions. It did not create new expression. The appeals court explicitly distinguished that design from generative AI, which produces new content, and from the copyright cases involving OpenAI, Anthropic and Meta.

That distinction limits the ruling’s reach: it resolves the use of Westlaw headnotes in this non-generative search tool, not the broader legality of training systems that generate original expression. Nicolas’s analysis cautions against reading the decision as either a categorical ban on AI training or blanket permission. The technology, source material, purpose and market effects still require case-specific scrutiny.

An appeal after the platform closed

Thomson Reuters sued in 2020. Ross shut down its platform in 2021, citing litigation costs, according to Reuters. The district court’s February 11, 2025 ruling found infringement and rejected fair use; Ross then appealed. The appellate victory therefore concerns a platform that had already stopped operating, rather than a newly imposed shutdown.