Third Circuit's ROSS ruling narrows fair use for AI training
The first appellate fair use decision on generative AI training reportedly went against ROSS Intelligence, leaving core AI-training copyright questions open across circuits.
On September 29, 2026, the U.S. Court of Appeals for the Third Circuit decided Thomson Reuters v. ROSS Intelligence, applying copyright fair use to the use of legal headnotes and related material in training generative AI. According to a King & Spalding client alert (the only source in this cluster; the full opinion was not available for review here), the panel ruled against ROSS and characterized the use as something other than "ordinary," leaving broader generative AI training questions unresolved.
Sophisticated counsel should care for three reasons. First, this is among the earliest appellate applications of the fair use doctrine to AI training, so its reasoning on the transformation and market-substitution factors will be cited in negotiations and motions nationwide. Second, a loss narrows the defensive playbook for AI developers, shifting emphasis toward licensed training data, output filtering, and indemnity structuring in vendor contracts. Third, the decision does not bind courts outside the Third Circuit, and its application to other model types or data sources remains unsettled, so it is a directional data point rather than a settled rule.
Practical next steps: read the opinion's factor analysis, map exposure for clients building or procuring legal-sector AI, and watch for en banc petitions and contrary rulings elsewhere.