Appeals Court Rules AI Training on Westlaw Headnotes Was Not Fair Use

A federal appeals court has ruled that a now-defunct legal tech startup broke copyright law when it used thousands of Westlaw headnotes to train an artificial intelligence search tool. In a decision filed Sept. 29, 2026, the U.S. Court of Appeals for the Third Circuit in Philadelphia affirmed a lower-court win for Thomson Reuters against ROSS Intelligence, rejecting ROSS’s argument that its AI training was fair use. Reuters described it as the first ruling by a U.S. appeals court in the wave of copyright cases over AI training.

The Ruling at a Glance

• Case: Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., No. 25-2153.
• Court: U.S. Court of Appeals for the Third Circuit. Judges Restrepo, Montgomery-Reeves and Bove heard the case; Judge Montgomery-Reeves wrote the opinion.
• Dates: argued June 11, 2026; decided Sept. 29, 2026.
• Holding: 2,243 Westlaw headnotes are protected by copyright, and ROSS’s use of them to train its AI was not fair use.
• Result: the court affirmed the District of Delaware’s Feb. 11, 2025, order granting Thomson Reuters partial summary judgment. Remaining issues go back to the trial court.

What Happened in the Case

Thomson Reuters, which owns the Westlaw legal research service and is the parent company of Reuters News, sued ROSS in 2020. Westlaw’s editors write “headnotes,” short summaries of the legal points in court opinions. According to the opinion and a summary by law firm Akin, a third party working for ROSS prepared about 25,000 legal memos that used Westlaw headnotes to frame legal questions. ROSS then used those memos to train its search engine to match questions with passages from court opinions.

ROSS shut down its platform in 2021, citing the cost of the litigation, Reuters reported. Judge Stephanos Bibas, a Third Circuit judge who heard the case at the trial level by special designation, rejected ROSS’s fair use defense in 2025, and ROSS appealed.

How the Court Applied Fair Use

Fair use is weighed using four factors. The court found three weighed against ROSS:

• Purpose and character: ROSS’s use was commercial; it planned to charge prices comparable to Westlaw’s. The court said training an AI was “arguably” a slight difference in use, but the ultimate purpose was the same as Westlaw’s, making it “minimally transformative, at best.”
• Amount used: ROSS copied the full text of the headnotes. Although that was only about 0.08% of Westlaw’s 28 million headnotes, the court said the copying was not necessary because the underlying court opinions are freely available.
• Market effect: the court found harm both to Westlaw and to a “rapidly developing” market for licensing headnotes as AI training data.

The second factor, the nature of the work, tilted slightly toward ROSS because headnotes are more factual than creative. Even so, the court found the headnotes had the “creative spark” needed for copyright protection. “In truth, this is no more than an ordinary copyright case,” Judge Montgomery-Reeves wrote.

Why It Matters for Generative AI

The ruling has limits. The court noted that ROSS’s AI “was not a generative AI” and could not produce original expression. Akin’s summary says the court distinguished generative AI cases, including Bartz v. Anthropic and the pending In re OpenAI Copyright Infringement Litigation. That means the decision does not settle whether training chatbots on books, news or images is fair use.

Still, the court’s market-harm reasoning, especially its recognition of a market for licensing training data, is likely to be cited by authors, publishers and other rights holders in the dozens of lawsuits against AI developers. Other AI content disputes we have covered include a voice actor’s lawsuit over an AI copy of his voice.

What It Means for Legal AI and Boston

Legal AI is a growing business, including in Boston, where legal AI firm Harvey has announced plans to hire in a new local office. The ruling signals that companies building legal tools need licenses for proprietary editorial content such as headnotes, even when the court opinions underneath are public. Regulators are also watching the AI industry closely, as our report on the FTC probe of OpenAI and Anthropic shows. More on our AI news hub.

Reactions

A Thomson Reuters spokesperson told Reuters 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.” Reuters reported that spokespeople and an attorney for ROSS did not immediately respond to requests for comment.

Featured photo: ajay_suresh / Wikimedia Commons, CC BY 2.0. Cropped and resized.

Sources

• U.S. Court of Appeals for the Third Circuit: Opinion, No. 25-2153 (Sept. 29, 2026)
• Third Circuit judgment, via CourtListener
• Reuters via Investing.com: US appeals court upholds Thomson Reuters’ landmark win in AI training lawsuit
• Akin: Third Circuit Finds AI Training Use of Westlaw Headnotes Is Not Fair Use
• GovInfo: Case record, USCOURTS-ca3-25-02153