The U.S. Court of Appeals for the Third Circuit has ruled for Thomson Reuters in its copyright suit against the now-shuttered AI legal research startup ROSS Intelligence. It is the first federal appellate decision to address fair use in the context of AI training.

The three-judge panel affirmed Judge Stephanos Bibas's February 2025 summary judgment on two core questions: whether Thomson Reuters's Westlaw headnotes qualify as copyrightable works, and whether ROSS's internal use of those headnotes as training data counted as fair use.

What did the Third Circuit actually decide?

The appellate panel upheld Judge Bibas's February 2025 rulings on both issues. The court found that the Westlaw headnotes were sufficiently original to qualify for copyright protection.

The court also found that ROSS's use of the material to train its system was not fair use. The opinion remains under seal while the court allows the parties to request redactions, so the full reasoning is not yet public.

What was the underlying dispute?

Thomson Reuters sued ROSS Intelligence, alleging that the startup had copied thousands of Westlaw headnotes — editor-written summaries of judicial holdings — to train a competing legal-search engine. The lower court granted summary judgment to Thomson Reuters on two questions:

• Whether a short quote or paraphrase of a judicial holding is copyrightable
• Whether fair use protected ROSS's internal use of Westlaw headnotes

The Third Circuit considered the case on an interlocutory appeal from that order.

Why is the ruling significant?

The decision is the first federal appellate ruling to address fair use in the AI training context. Lawyers tracking the AI copyright space have followed the ROSS case closely since Judge Bibas ruled in February 2025.

Some observers have suggested the case could influence pending litigation involving OpenAI, Anthropic, and Meta over their use of copyrighted materials to train their own models.

But the unique facts may limit the ruling's reach. ROSS's copying occurred before generative AI's widespread popularity and was for a non-generative system. Judge Bibas also found that ROSS was building a product designed to directly compete with Thomson Reuters — a finding that could distinguish the case from suits against general-purpose AI chatbots. ROSS has since ceased operations.

What did the parties say?

A Thomson Reuters spokesperson told Reuters that the company was "pleased with the ruling" and "firmly believes that respecting copyright is essential for fostering innovation while protecting the intellectual property that fuels fiduciary-grade AI solutions."

What happens next for practitioners?

The Third Circuit's opinion is sealed while the parties may request redactions. Practitioners should watch for the unsealed opinion, since the court's reasoning will determine the case's reach.

An affirmance that closely follows the facts could be consequential but fairly confined to the Westlaw headnote context. A broader rule could reverberate across the AI industry and reshape how courts evaluate fair use in machine-learning disputes.

The case reached the Third Circuit as an interlocutory appeal, so further proceedings remain possible after the mandate issues. For now, the ruling leaves Judge Bibas's two core holdings undisturbed: the headnotes are copyrightable, and ROSS's training use was not fair use. Attorneys advising AI developers and content owners will study the appellate reasoning once unsealed to gauge how broadly the precedent extends beyond headnotes.