Third Circuit rejects fair-use defence in Thomson Reuters v. ROSS, first US appeals ruling on AI-training copyright

What happened: On 2026-09-29, the US Court of Appeals for the Third Circuit affirmed that ROSS Intelligence infringed 2,243 Thomson Reuters (Westlaw) headnotes by copying them into AI training memos for a competing legal-research tool, and rejected ROSS's fair-use defence as a matter of law. The court held ROSS's copying was 'minimally transformative, at best' — 'Unlike necessity, ease is not a justification for copying' — and found that AI-training-data licensing is a real, rapidly developing market ROSS usurped. The opinion expressly sets generative AI aside in a footnote, responding to the DOJ's statement of interest in the OpenAI litigation. ROSS says it will seek Supreme Court review; damages and remaining headnote claims go to trial in Delaware.

Why it matters to creators: this is the first federal appellate ruling on whether AI-training copying is fair use. It establishes that a licensing market for AI training data exists (fourth fair-use factor weighs against unlicensed trainers) — a precedent rightsholders can cite in the parallel generative-AI cases. But the court deliberately carved generative LLMs out, so it does not decide the OpenAI/Anthropic cases; commentators note it may set up a circuit split with Bartz v. Anthropic.

Not yet known: whether the Supreme Court takes the case; the Delaware damages-trial outcome and which headnote copyrights have expired.