US appeals court upholds Thomson Reuters win in AI training lawsuit

A US federal appeals court on Tuesday upheld a ruling in favour of Thomson Reuters in its long-running copyright dispute with Ross Intelligence, rejecting the defunct legal-research company's argument that its use of Westlaw material to train an AI search engine was fair use.

The ruling

The Philadelphia-based 3rd US Circuit Court of Appeals affirmed the earlier decision, Reuters reported. It is the first time a US appeals court has ruled on a copyright case over AI training, though the panel's full reasoning is currently sealed. Thomson Reuters, which brought the case, is the parent company of Reuters News.

What Ross copied

Thomson Reuters sued Ross Intelligence in 2020, accusing the company of copying thousands of "headnotes" from its Westlaw platform. Headnotes are short editorial summaries of the points of law set out in court opinions. According to the complaint, Ross used that material to train a competing AI-based legal search engine. Ross shut down its platform in 2021, citing the cost of the litigation.

Ross argued that its AI training amounted to fair use of the headnotes, a defence that has become central to disputes over how AI systems are built. A federal judge in Delaware rejected that argument last year, prompting the appeal. "Ross took the headnotes to make it easier to develop a competing legal research tool," the judge wrote. "So Ross's use is not transformative."

Why it matters

The case sits inside a larger wave of litigation. Authors, news outlets and music labels have filed dozens of lawsuits against technology companies over the use of copyrighted works to train AI systems. Rights holders say the companies turn their work into competing content, while technology firms argue that generative AI transforms the material into something new.

The Thomson Reuters dispute differs from most of those cases because it does not involve generative AI. Ross used the headnotes so that its search engine could return relevant court opinions in response to user queries, rather than to produce new text.