Source: ArentFox Schiff

The US Court of Appeals for the Third Circuit affirmed copyright and fair use determinations concerning ROSS Intelligence’s non-generative AI legal research tool, according to professional commentary by Danielle W. Bulger and Perry A. Jackson. Thomson Reuters alleged that ROSS used Westlaw’s editorial headnotes to develop training materials; ROSS obtained those materials through LegalEase after Thomson Reuters declined to license its content. The district court granted partial summary judgment concerning 2,243 headnotes, while leaving issues involving thousands of other headnotes and possible copyright expiration for trial. In an opinion by Judge Tamika Montgomery-Reeves, the appellate court reviewed originality and fair use through an interlocutory appeal. The court found sufficient originality in editors’ selection and expression of legal points, while explaining that judicial opinions remained freely available. The court did not resolve protection for headnotes quoting opinions verbatim or examine the West Key Number System’s originality, which ROSS did not challenge on appeal.

The court found the AI training use commercial and only minimally transformative because both platforms used the headnotes for legal research. The court rejected ROSS’s reliance on intermediate copying cases involving computer code, reasoning that ROSS could have used the underlying judicial opinions. Copying complete headnotes without a transformative purpose also weighed against fair use, alongside harm to legal research competition and the developing market for licensing headnotes as AI training data. The first, third and fourth fair use factors weighed against ROSS; only the second weighed slightly in ROSS’s favor. The court also noted evidence of access contrary to Westlaw’s terms and treated that evidence as adverse to ROSS insofar as good faith remained relevant. The ruling expressly distinguished generative AI and emphasized that ROSS’s tool could not create original expression and was designed to replace Westlaw. Bulger and Jackson suggest that the reasoning might influence disputes involving specialized models, while stressing that federal appellate courts had not resolved fair use for generative AI training.