In Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., No. 25-2153, a Third Circuit panel (Judges Restrepo, Montgomery-Reeves and Bove) affirmed a District of Delaware order granting Thomson Reuters partial summary judgment. Judge Montgomery-Reeves wrote the opinion, filed 29 September 2026. The appeal was interlocutory: the district judge, Third Circuit Judge Stephanos Bibas sitting by designation, had certified the copyrightability and fair-use questions for review.

The facts the court relied on

ROSS built an AI legal search engine that answered plain-language questions with passages from about ten million public judicial opinions. To train it, ROSS hired LegalEase Solutions, which wrote about 25,000 training memos. Their drafters used thousands of Westlaw headnotes to frame the legal questions. The court noted that ROSS’s AI “was not a generative AI”: it “would only return text passages from preexisting judicial opinions.”

What the court held

  • Copyright: the 2,243 headnotes at issue are original enough to be protected.
  • Purpose and character (factor 1): against ROSS. Its use was highly commercial and “minimally transformative, at best,” because it served the same purpose as Westlaw’s: a legal-research platform.
  • Nature of the work (factor 2): slightly favours ROSS.
  • Amount used (factor 3): against ROSS.
  • Market effect (factor 4): against ROSS. ROSS aimed to be a direct substitute, and the court found harm in a “rapidly developing” market for licensing headnotes as AI training data. That Thomson Reuters had not licensed them to others “does not disprove that a market exists.”
“In truth, this is no more than an ordinary copyright case.” — Third Circuit, Thomson Reuters v. ROSS, No. 25-2153, 29 September 2026

What it does not decide

In a footnote, the court noted a Department of Justice statement of interest filed on 1 September 2026 in In re: OpenAI, Inc. Copyright Infringement Litigation (S.D.N.Y.), which argues that training a large language model that can generate original responses is transformative. The panel said those concerns “do not apply here,” because ROSS’s platform cannot generate original expression and was built as a commercial substitute for Westlaw. Whether training generative models on books, news or art is fair use remains open.

Citizen impact: This is the first time a federal appeals court has applied fair use to AI training. It tells AI developers in Pennsylvania, New Jersey, Delaware and the U.S. Virgin Islands, and signals to courts elsewhere, that copying protected material to build a product that competes with the original, without a licence, is not shielded just because the copying feeds an AI. It also treats the market for licensing works as AI training data as real, which strengthens the hand of publishers seeking licence fees. It does not settle the cases against generative-AI companies.

What this piece refuses

Do not write that “AI training is now illegal” or that the ruling decides the OpenAI, Meta or Anthropic cases. It concerns a non-generative search tool and binds only the Third Circuit. Do not describe the DOJ statement as the government’s position in this case: the court noted that DOJ did not file here.

Bottom line

Verified from the Third Circuit’s opinion in No. 25-2153 (filed 29 September 2026): Westlaw headnotes are copyrightable, and ROSS’s use of 2,243 of them to train a competing AI search tool was not fair use. Affirmed. The generative-AI question is expressly left open.