AI training of copyrighted material not fair use: Third Circuit

Thomson Reuters sued ROSS Intelligence after ROSS obtained Westlaw headnotes through a third party and used them to train an AI that competed with Westlaw’s research product. A federal judge found the headnotes copyrightable because their selection and arrangement added a “creative spark” beyond the underlying judicial opinions, and ruled ROSS’ unlicensed copying violated copyright. The court concluded the headnotes were original works and that ROSS’ use was insufficiently transformative, since the startup employed the material to develop a competing legal-research platform rather than for a new expressive purpose.

A Third Circuit panel, in an opinion by Judge Tamika Montgomery-Reeves with Judges Restrepo and Bove concurring, affirmed those rulings. The appeals court reiterated that the headnotes meet the minimal originality threshold, described ROSS’ AI training as only minimally transformative, and held that the copying harmed both Westlaw’s market and potential derivative licensing markets for training data - undermining a fair-use defense. ROSS says it will seek further review, arguing the decision leaves uncertainty around copyright’s application to AI training, while Thomson Reuters hailed the ruling. The outcome may become a significant precedent limiting unlicensed use of copyrighted material for AI model training.