A US appeals court upheld that training AI on a rival's content wasn't fair use
On 30th Sep'26, the US Court of Appeals for the Third Circuit affirmed that Ross Intelligence infringed Thomson Reuters' copyright by using more than 2,000 Westlaw headnotes to build its AI legal research tool, and that fair use did not protect it. It is the first US appeals court ruling on fair use for AI training. The opinion is sealed while the parties agree redactions, the case involved a non-generative search tool, and Ross says it will ask the Supreme Court to hear it.
What it actually means
Until the reasoning is published, read this through the facts the court affirmed rather than the headlines about it. Ross built a product meant to compete with Westlaw from Westlaw's own material, and the lower court found that competition was what sank the fair use defence. That is the pattern to check in your own build: if you are training or fine-tuning on a particular source's content, and your product would do the job that source sells, you are much closer to Ross than to a general model trained on the open web.
For most operators the practical step is a short list of where your training and fine-tuning data came from, with the licence beside each source. The ruling does not settle the big generative cases, and the Supreme Court may yet take it, but procurement and legal teams will now ask the question, and it is easier to answer with that list in hand.
If your product does the job your training data's owner sells, you're standing where Ross stood.
Source: LawNext · Sunday 4 October 2026