US Third Circuit: AI training not fair use after all

The hot news in AI training is a development in a case from before chatbots as we know them — Thomson Reuters v. Ross Intelligence from 2020. Thomson Reuters owns Westlaw, which publishes legal opinions with summaries. The summaries are copyrighted. Ross was training a machine learning model on the…

The hot news in AI training is a development in a case from before chatbots as we know them — Thomson Reuters v. Ross Intelligence from 2020. Thomson Reuters owns Westlaw, which publishes legal opinions with summaries. The summaries are copyrighted. Ross was training a machine learning model on the Westlaw summary text. Thomson Reuters sued in 2020 and won in 2025. The bankrupt shell of Ross Intelligence appealed to the Third Circuit — who rejected the appeal on 29 September. [Courthouse News] The generative AI industry has been relying on the 2025 ruling in the Anthropic books case that training a chatbot was fair use if you’d bought the books. They’re wondering if the new Ross ruling could put that into question. The judges here held that Anthropic does not apply in the Ross case. This Ross ruling shouldn’t be huge news — the original verdict against them came before the Anthropic ruling — but it’s been a bit of a shock to an AI industry that thought they were in the clear. Judge Tamika Montgomery-Reeves wrote: [opinion, PDF, archive] Under ROSS’s framing, this case appears to concern the future of AI legal technology. But appearances can be deceiving. … The undisputed evidence reflects that Thomson Reuters’s materials possess a creative spark, and ROSS aspired to be a direct competitor by using them for a highly similar purpose. We therefore hold that Thomson Reuters’s materials are copyrightable and that ROSS’s use was not fair. Fair use rulings are very dependent on the particular facts and circumstances, so this new ruling doesn’t promptly blow up the Anthropic ruling. But it does raise all sorts of exciting legal questions! Fair use in the US has various factors in law. How these apply is different in every case. Feeding the Common Crawl of the whole Internet into your research model is different to feeding it to Claude, and that’s different to feeding Westlaw into Ross Intelligence’s machine learning model. Ross Intelligence plans to take this appeal to the Supreme Court. I have no idea how the Supreme Court would rule on it, or if they’d even hear the case. If they did, I’d expect amicus briefs from every AI company about how they should let Ross Intelligence off. If the Supreme Court did let Ross off, what would happen next? Anthropic still have that previous ruling. But someone bringing a new case against Anthropic could use a (hypothetical) ruling in Ross’s favour. If that happened, we might end up with the AI vendors finally being forced to pay licenses for everything. Like Anthropic wasn’t in enough financial deep water. But this is all speculative right now. You probably can’t use this Ross appeal rejection as a fresh bludgeon for Anthropic. Yet. Probably nothing much will happen. But it’ll definitely be fun! Video — Podcast

Source: Pivot to AI — Published October 16, 2026 at 20:52 UTC — Category: Business