The Publishing Times

All the News That’s Fit to Publish

publishing-news · Guides

Third Circuit: Training AI on Westlaw Headnotes Was Not Fair Use

The Third Circuit affirmed that ROSS Intelligence's copying of 2,243 Westlaw headnotes to train a competing legal-search AI was not fair use.

Can an AI company copy a publisher's editorial headnotes to train a competing search tool and call it fair use? The U.S. Court of Appeals for the Third Circuit says no. In an opinion filed September 29, 2026, and released September 30, a three-judge panel affirmed partial summary judgment for Thomson Reuters and West Publishing. The court held that 2,243 Westlaw headnotes are original enough for copyright, and that ROSS Intelligence's use of those headnotes to train a competing AI legal-research platform was not fair use. The Association of American Publishers welcomed the decision the next day, calling it a milestone for rightsholders in the AI market.

What the court decided

Thomson Reuters sued ROSS in the District of Delaware in May 2020. ROSS had wanted access to Westlaw. When Westlaw refused, ROSS used a third party to create AI training memos that copied thousands of Westlaw headnotes — short editorial summaries of legal points that appear before judicial opinions — so ROSS could train a legal-search tool meant to compete with Westlaw.

Judge Stephanos Bibas, sitting by designation, granted Thomson Reuters partial summary judgment in February 2025 on originality, infringement, and fair use for those 2,243 headnotes. He certified two questions for immediate appeal: whether the headnotes are original as a matter of law, and whether ROSS's use was fair. The Third Circuit answered both for Thomson Reuters.

Writing for the panel (Judges Restrepo, Montgomery-Reeves, and Bove), Judge Tamika Montgomery-Reeves framed the appeal as "no more than an ordinary copyright case." The court held that Thomson Reuters's materials possess a creative spark, and that ROSS "aspired to be a direct competitor by using them for a highly similar purpose." On that record, the court affirmed.

Why the fair-use defense failed

ROSS never showed a headnote to its end users. The finished tool returned passages from judicial opinions. ROSS argued that intermediate copying for AI training should be fair use, pointing to cases where copying was necessary to achieve a different purpose.

The Third Circuit disagreed. It found the use highly commercial and not transformative in the sense that mattered: ROSS used the headnotes to help users find judicial opinions related to legal research inquiries — something Thomson Reuters already does with those headnotes. The panel wrote that, unlike necessity, "ease is not a justification for copying."

On the market-effect factor, the court said the market for licensing headnotes as text to train AI is "rapidly developing," and that Thomson Reuters presented evidence it is using its own headnotes as training data for its AI search products. That Thomson Reuters had not licensed the headnotes to others, the court said, does not disprove that a market exists. The first, third, and fourth fair-use factors weighed against ROSS; the second weighed only slightly in favor. Taken together, the use was not fair.

What remains for trial in Delaware includes expiration questions on some copyrights, claims on remaining headnotes, and damages. ROSS is now defunct; the appeal concerned originality and fair use.

How publishers read the decision

In its October 1, 2026 statement, AAP President and CEO Maria A. Pallante said rightsholders "are entitled to equitable participation in the AI market," and that the court upheld "the basic principles of fairness that will be central to a future in which innovation and creativity work together for all stakeholders and the public." She noted the court's point that AI is not the first technology to exploit expressive works, and that ROSS "usurped Thomson Reuters's right to enter a derivative market for licensing headnotes for AI training."

AAP had filed an amicus brief in November 2025 asking the Third Circuit to affirm. In the statement, Pallante repeated the brief's argument that fair use cannot "ignore markets to which rightsholders are fundamentally entitled and are both exploiting and preparing to exploit." She said the court correctly noted that the licensing market for AI training not only exists but is rapidly developing, and that the court rejected the idea that a rightsholder's decision not to license disproves the market. "We applaud the court for this powerful precedent," she said, "which supports the premise that a robust licensing market among developers and rightsholders is essential to achieving sustainable and ethical progress in AI technology."

AAP also hosts a copy of the Third Circuit opinion with its statement.

Why it matters for authors

This is the first federal appellate decision to hold that training a commercial AI system on a publisher's copyrighted editorial content, without a license, was not fair use — at least on these facts. The product was a narrow legal-search tool that aimed to replace Westlaw, not a generative model that outputs new prose. The panel itself treated the dispute as an ordinary copyright case about competing with the work you copied.

Authors should still read the holding carefully. The court leaned on direct competition, commercial purpose, substantial copying of creative headnotes, and a developing licensing market for AI training data. Other AI cases — including generative-model suits that draw different lines around training versus piracy — are still moving. What this opinion puts on the record is that "we trained on it and did not show the user the original" is not, by itself, a fair-use shield when the tool is built to take the rightsholder's place and a licensing market for that training use is forming.

For authors watching settlements and licensing talks, the takeaway is sharper: courts will look at whether the AI product substitutes for the copyrighted work and whether a market to license training access exists. On that map, the Third Circuit drew a clear line against competitor training on editorial headnotes.

Sources