Thomson Reuters v. ROSS: the Third Circuit holds that ROSS's copying of Westlaw headnotes to train its legal AI tool was not fair use

What the court decided

A Third Circuit panel of Judges Restrepo, Montgomery-Reeves, and Bove heard argument on June 11, 2026, and filed its opinion on September 29, 2026. Judge Montgomery-Reeves wrote for the court. The appeal came from the District of Delaware, No. 1:20-cv-00613, where Circuit Judge Stephanos Bibas presided by designation. On February 11, 2025, he granted Thomson Reuters partial summary judgment, holding that ROSS's use of 2,243 Westlaw headnotes infringed and was not fair use, and he then certified two questions for interlocutory appeal at the parties' request.

The panel affirmed. In its words: "We therefore hold that Thomson Reuters's materials are copyrightable and that ROSS's use was not fair." The certified questions also asked about the originality of West's Key Number System, but ROSS never briefed that issue, so the court treated it as forfeited and did not decide it.

The facts the court relied on

ROSS built an AI legal search engine that answered plain-language questions with passages from judicial opinions. The court describes the system this way: "ROSS's AI was not a generative AI, meaning it would not create any new expression; it would only return text passages from preexisting judicial opinions."

To train it, ROSS enlisted LegalEase Solutions, which wrote approximately 25,000 memos, each pairing a legal question with four to six opinion passages. According to the opinion, the memo drafters used thousands of Westlaw headnotes to frame the questions. On appeal, ROSS did not dispute that LegalEase's work is attributable to ROSS or that the drafters copied headnotes. The district court limited its ruling to 2,243 headnotes where the memo questions were so close to the headnote text, and so far from the opinion text, that no reasonable juror could find they were not copied. The opinion also notes that ROSS advertised its platform against Westlaw at prices "in line with" Westlaw, and that some law firms switched.

Originality: headnotes have a creative spark

Applying the low originality bar of Feist Publications, Inc. v. Rural Telephone Service Co., the court held: "All 2,243 independently created headnotes at issue in this appeal have 'some creative spark.'" Editors decide which points of law matter and how to state them so each headnote stands on its own. The court relied on Callaghan v. Myers (1888), which recognized a case reporter's copyright in headnotes the reporter wrote, and on Georgia v. Public.Resource.Org, Inc. (2020), which described that rule.

It rejected ROSS's three counterarguments. On the claim that a copyright in headnotes would give Thomson Reuters a monopoly over the law, the court said: "Headnotes are not law; judicial opinions are." The merger doctrine did not apply because, as the district court had said and the panel agreed, "there are many ways to express points of law from judicial opinions." And Matthew Bender & Co. v. West Publishing Co. (2d Cir. 1998) involved party names and parallel citations, not independently composed headnotes. The court expressly left open whether a headnote that quotes an opinion word for word could be original, because none of the 2,243 headnotes did.

Fair use, factor by factor

• Purpose and character: against fair use. ROSS's use was commercial. Training an AI program was an intermediate step that "arguably presents a slight degree of difference in use," but both companies used the headnotes to build legal-research platforms, which made ROSS's use "minimally transformative, at best."
• Nature of the work: favors fair use. The headnotes were published and are more factual than fictional. The court's conclusion says this factor weighs slightly in favor of fair use.
• Amount used: against fair use. ROSS argued it took only 0.08 percent of Thomson Reuters's 28 million headnotes. The court answered that copying was not necessary because the opinions themselves were freely available, and it affirmed that each headnote is its own copyrightable work.
• Market effect: against fair use. The court concluded that ROSS harmed the headnotes' value as a draw for Westlaw subscribers. It reasoned that unrestricted and widespread copying like ROSS's would harm Westlaw's position in the legal-research platform market, a harm ROSS offered no evidence to rebut. And it held that ROSS harmed a potential market for licensing headnotes as AI training data, which it said "is rapidly developing."

Weighing the four factors together, the court held that ROSS's use was not fair.

Why the intermediate-copying cases did not help ROSS

ROSS relied on Authors Guild v. Google, Inc. (2d Cir. 2015), Google LLC v. Oracle America, Inc. (2021), Sega Enterprises Ltd. v. Accolade, Inc. (9th Cir. 1992), and Sony Computer Entertainment, Inc. v. Connectix Corp. (9th Cir. 2000). The court distinguished all of them. Google Books served a different function and could lead readers to buy the books, while "ROSS's use does not lead a user to Westlaw; instead, as ROSS admits, it aims to replace Westlaw." The software cases involved copying that was necessary to reach unprotected functional elements. ROSS had the opinions and chose headnotes because they were an easy way to frame questions. "Unlike necessity, ease is not a justification for copying."

In a footnote, the court added that the undisputed evidence "reflects that ROSS at times acted in bad faith," citing record evidence about how ROSS personnel tried to access Westlaw, and that to the extent good faith still matters, it weighs against fair use.

What the opinion says about generative AI

Footnote 7 addresses the U.S. Department of Justice's September 1, 2026 statement of interest in In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-3143 (S.D.N.Y.). The court said: "The concerns raised in that separate case do not apply here." It noted that the DOJ relied on Bartz v. Anthropic PBC to argue that training a large language model that can generate original responses is transformative, and that, unlike the models in those cases, ROSS's platform "cannot generate original expression." The opinion therefore does not decide whether training a generative model is fair use.

The court also rejected ROSS's public-benefit arguments. The opinions are already free to read, ROSS charged prices comparable to Westlaw, ROSS offered no evidence that the ruling would halt AI development, and it offered no evidence tying its platform to national security. The court added that while some AI technology may implicate national security concerns, "this does not give ROSS carte blanche to violate copyright law merely because it incorporates AI."

What it means for AI developers and content owners

The court held that, where the underlying opinions were freely available, the ease of copying Thomson Reuters's headnotes did not justify it, and that ROSS's effort to build a substitute for Westlaw weighed against fair use. It also counted licensing headnotes as AI training data as a potential derivative market, on evidence that this market "is rapidly developing," even though Thomson Reuters had not licensed its headnotes to others. The opinion notes that Thomson Reuters uses its headnotes to train its own AI search products.

The ruling resolves the certified appeal from the partial summary judgment order. It does not decide whether training a generative AI model is fair use, and it does not decide whether West's Key Number System is original.