Thomson Reuters v. Ross
The first US ruling to reject fair use as a defence for training an AI on copyrighted material.
Ross Intelligence built an AI legal research tool and trained it on question-and-answer memos that a contractor had derived from Westlaw's headnotes, the short summaries Thomson Reuters attaches to each decision. Thomson Reuters sued in 2020. Ross shut down the following year, citing the cost of the litigation.
In February 2025 Judge Stephanos Bibas found the copying infringing and the fair-use defence wanting, having leaned the other way himself in 2023. What tipped it was that Ross was building a direct competitor to the work it copied. Bibas noted explicitly that the system was not generative, which limits how far the ruling reaches in the cases against language models – but every party to those cases has read it.