AI training of copyrighted material not fair use: Third Circuit

Circuit Court of Appeals ruled that using Thomson Reuters’ copyrighted legal headnotes to train artificial‑intelligence models does not qualify as fair use. The decision marks a landmark win for the publisher and a setback for startups that rely on large datasets of legal material. The dispute began when Reuters sued legal‑research startup ROSS for incorporating its proprietary headnotes into a generative‑AI product.

The case advanced through the district court before reaching the appellate panel, which issued an unsealed opinion detailing why the copying exceeded permissible limits. Coverage in Reuters and Nieman Journalism Lab described the ruling as a “landmark AI copyright win,” while lawnext.com noted the victory for Reuters in the broader fight over data rights. With the appellate decision now precedent, AI developers must avoid training on Reuters’ protected legal content unless a license is secured.

Industry observers note the decision could affect how training data is sourced across sectors. The ruling signals that courts may scrutinize other copyrighted datasets used for machine‑learning, a point highlighted by PYMNTS.com.