Thomson Reuters v Ross: the US Court of Appeals for the Third Circuit confirms that training an AI on Westlaw summaries did not constitute fair use

For the first time, a US federal court of appeal has ruled on a case concerning the training of AI using copyright-protected content, and has found in favour of the publisher.

On September 29, the Federal Court of Appeals for the Third Circuit upheld Thomson Reuters' victory over Ross Intelligence. Since 2020, the group had criticized the start-up for training its legal search engine on Westlaw's "headnotes", these short summaries of key points in court decisions, obtained through a licensee to build a competing product.

The judges considered that these summaries were well protected by copyright and that their use to train Ross's tool did not fall under fair use. Particularity: the judgment was initially handed down under seal, its detailed reasons not being immediately accessible to the public.

Limited scope to non-generative AI

This is the first time that a US appellate court has ruled on the right of AI companies to train on publishers' content, while dozens of lawsuits have targeted OpenAI, Anthropic, Meta and Stability AI. Commentators point out, however, that Ross did not produce text: the engine linked to existing decisions. The question of fair use for generative models therefore remains open.