On Tuesday, a U.S. federal appeals court ruled in favor of Thomson Reuters, upholding the company’s landmark win in a copyright infringement case against the now-defunct legal AI startup Ross Intelligence. The ruling marks the first federal appeals case to reject a fair use argument for using publisher content, without permission or compensation, to train a commercial AI product.
“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,” wrote Judge Tamika Montgomery-Reeves, who sits on the 3rd U.S. Circuit Court of Appeals in Philadelphia, in her decision.
Notably, Montgomery-Reeves points out that the case does not concern a large-language model (LLM) or other generative AI tool. Ross’s competing AI search engine didn’t generate original text, but instead retrieved existing passages from judicial opinions. She argues that this may limit its impact on other AI copyright litigation.
“Under ROSS’s framing, this case appears to concern the future of AI legal technology. But appearances can be deceiving,” she writes. “In truth, this is no more than an ordinary copyright case.”
Thomson Reuters filed its lawsuit back in 2020, long before the release of ChatGPT and other generative AI chatbots as we know them. The first known AI copyright lawsuit in the U.S., the case echoes many more recent lawsuits between news publishers and AI companies like OpenAI and Anthropic. Thomson Reuters argued that Ross had copied thousands of “headnotes,” short summaries of the legal issues in a given case, that were published on its legal research platform Westlaw. These headnotes were then used to train Ross’s own competing AI-powered legal search engine.
In a February 2025 ruling, judge Stephanos Bibas sided with Thomson Reuters in the case. He rejected the argument that using copyrighted work to train the commercial AI model was legally permissible, as Ross had argued in court. One of the main factors in a fair use case is whether the infringing work impacts the market value of the original.
“Even taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute,” wrote Bibas in his original ruling. “And it does not matter whether Thomson Reuters has used the data to train its own legal search tools; the effect on a potential market for AI training data is enough.”
Legal experts have been keeping a close eye on Ross’s appeal with the 3rd U.S Circuit Court of Appeal. As the first AI copyright case to reach a federal appeals court, there was potential for it to set far-reaching legal precedent on fair use — precedent that could be taken advantage of by news publishers arguing in similar AI copyright cases.
Ultimately, Montgomery-Reeves’ ruling emphasizes that an AI tool that reproduces a copyrighted work’s text in order to compete directly with that work in the market, will likely not hold up to a fair use test. But she leaves room for generative AI tools to be considered more transformative than Ross’s AI search engine.
In the lawsuit Bartz vs. Anthropic, filed by members of the Authors Guild, a judge ruled that Anthropic’s use of published books to train its Claude LLMs was, in fact, fair use. It was only Anthropic’s decision to pirate books and store copies of them that lost them in that case. That ruling in June 2025 resulted in the largest settlement in U.S. copyright law history, granting a $1.5 billion payout to authors and publishers.
“Unlike the AI models in Bartz…ROSS’s AI platform cannot generate original expression, and the evidence here supports the opposite conclusion about transformativeness,” wrote Montgomery-Reeves, implying that her ruling does not contradict the ruling in that case.
In other words, if Ross had built a product that could output summaries of Westlaw’s headnotes with its own “original expression,” this fair use ruling could have looked very different.
