Judge Stephanos Bibas of the 3rd U.S. Circuit Court of Appeals, sitting by designation, ruled that Ross Intelligence cannot claim fair use after lifting thousands of Westlaw headnotes to train an AI legal assistant. The summary judgment order, issued Tuesday, hands Thomson Reuters a pivotal win in one of the earliest copyright fights over AI training data.

Ross shut down in 2021, but Thomson Reuters pressed forward with its 2020 lawsuit, arguing the startup replicated the structure and wording of its Westlaw editorial summaries to bootstrap a rival research product. Bibas agreed, writing that Ross wanted “to compete head-to-head with Westlaw” and that none of its defenses “holds water” once the copying was viewed through that commercial lens.

The decision arrives as publishers, novelists, and record labels take OpenAI, Stability AI, and other model developers to court over the practice of ingesting copyrighted works without licenses. Thomson Reuters’ victory offers the clearest signal yet that courts may not treat wholesale scraping of proprietary databases as transformative just because the end product is AI-powered.

While the damages phase will continue, the ruling jolts legal tech vendors that rely on outside datasets to train retrieval and drafting agents. Law firms told SeedAI News they expect renewed pressure to document provenance for every model fine-tune, especially when headnotes, case briefs, or treatises enter the mix.

Microsoft and OpenAI, which face separate author and newsroom lawsuits, are monitoring the outcome closely. If judges follow Bibas’ reasoning, large model operators could be forced into broad licensing deals or risk losing the safe harbor that tech companies have leaned on during the AI boom.