Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc.

Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., decided February 11th, 2025. Judge Stephanos Bibas, a Third Circuit judge sitting by designation in the District of Delaware, granted partial summary judgment for Thomson Reuters. He held that Westlaw's headnotes and Key Number System are original and copyrightable, that Ross copied protected expression, and that Ross's use of that material to train a competing legal-research tool was not fair use. It is the first major American decision on copyrighted material and AI training. Here's the brief.

Thomson Reuters owns Westlaw, one of the largest legal-research platforms. Alongside case law and statutes, Westlaw carries editorial content: headnotes, which distill the key points of law from a judicial opinion, and the Key Number System, a numerical taxonomy that organizes the law by topic.

Ross Intelligence was a newer competitor building a legal-research tool that used artificial intelligence. To train it, Ross needed a database of legal questions and answers, so it asked to license Westlaw's content. Because Ross was a competitor, Thomson Reuters refused.

Ross then turned to a company called LegalEase, which supplied roughly twenty-five thousand "Bulk Memos" — compilations of legal questions with good and bad answers, created with the help of Westlaw headnotes. Ross used the Bulk Memos to train its AI. When Thomson Reuters found out, it sued for copyright infringement.

In 2023, Judge Bibas had largely denied summary judgment, and the case moved toward trial. Studying the record more closely, he concluded that his earlier ruling had not gone far enough, and he invited the parties to renew their briefing. This February 2025 opinion revises that 2023 decision.

The core dispute was whether the Bulk Memo questions copied protected expression in Thomson Reuters's headnotes, or instead drew only on the uncopyrightable text of the judicial opinions — and, if Ross copied, whether its use was fair.

Copyright protects only original works, but the originality threshold is low: under Feist, a work needs just some minimal degree of creativity. Fair use is an affirmative defense, on which Ross bore the burden, governed by a four-factor statutory test.

The court noted that the first and fourth factors weigh most heavily, and that on these undisputed facts fair use was primarily a question of law for the judge.

On copyrightability, the court found no genuine dispute that the headnotes and Key Number System clear Feist's minimal threshold for originality. A headnote, the court reasoned, is a point of law chiseled out of a longer opinion — like a sculptor choosing what to cut away — and that editorial judgment supplies the required creative spark. Comparing the materials one by one, the court found actual copying of 2,243 headnotes and granted summary judgment to Thomson Reuters on infringement of those headnotes, subject to remaining copyright-validity questions for a jury.

The court then rejected Ross's fair-use defense. On factor one, Ross's use was commercial and not transformative.

The court distinguished cases permitting intermediate copying of computer code, finding that reasoning inapt where the copied material was written words, not code. Factors two and three favored Ross — the works were not highly creative, and Ross's end product did not deliver headnotes to users — but the court gave them less weight. Factor four, the most important, went to Thomson Reuters: Ross had built a market substitute, and its copying threatened a potential market for AI training data.

The court was careful about the limits of its ruling. It stressed that Ross's tool was not generative AI: rather than writing new text, it returned judicial opinions that had already been written.

So the decision resolves fair use for one non-generative tool trained on copyrighted material; it does not decide the question for generative AI. As the first major ruling on copyright and AI training, it stands for the proposition that copying protected expression to build a competing product is not automatically fair use. The case is now on appeal to the Third Circuit.

Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., 765 F. Supp. 3d 382, decided February 11th, 2025. I'm John Goodhue. Thanks for watching.