Summary
The U.S. Court of Appeals for the Third Circuit on Tuesday became the first federal appellate court to decide whether using copyrighted material to train an AI system qualified as fair use. In ruling for Thomson Reuters, the court held that headnotes in the company’s Westlaw legal intelligence platform are sufficiently original and, thus, protected by copyright, and that use of those headnotes to train a product did not qualify as fair use by legal research firm ROSS Intelligence.
The Upshot
- ROSS Intelligence used the headnotes to train an AI product that returned passages from existing judicial opinions and was designed to compete directly with Westlaw. The court concluded in its September 29 ruling that the use was highly commercial and only minimally transformative. It also found harm to both Westlaw’s existing legal-research market and a developing market for licensing headnotes as AI-training data.
- The Third Circuit’s decision marks the first federal appellate ruling on fair use in AI training. But the reach of the opinion is limited because the AI technology at the heart of the case was itself relatively limited: ROSS Intelligence’s AI product used training data to identify and return existing judicial passages rather than generate new expression.
- The first fair-use factor, which examines the purpose and character of the challenged use, favored Thomson Reuters. The Third Circuit held that ROSS Intelligence’s use of the headnotes was only minimally transformative because its AI product used the headnotes in essentially the same way that Westlaw was using the headnotes: to improve a legal-research service.
- The fourth fair-use factor, which examines market harm, also favored Thomson Reuters. The court identified harm in two distinct markets: the existing legal-research market, where ROSS Intelligence’s AI product was designed to compete directly with Westlaw, and a developing market for licensing headnotes as AI-training data.
- The Third Circuit distinguished ROSS Intelligence’s AI product from systems that can generate original expression, leaving the central generative-AI debate unresolved. The opinion does not decide whether training a generative AI model on copyrighted books or other expressive works is fair use.
The Bottom Line
Thomson Reuters v. ROSS Intelligence Inc. is the first federal appellate decision to address fair use in AI training. But the opinion does not answer the larger questions surrounding generative AI because the technology before the court did not generate new expression; it simply used Westlaw headnotes to improve a legal-research product designed to compete directly with Westlaw. The court’s explicit clarification that these facts differed from the pending generative AI cases means that more complex questions, such as whether using copyrighted works to train generative AI platforms can be fair use, remain open. When a generative model uses copyrighted works to learn patterns that later allow it to create new text, images, music, or code, courts still must decide how to weigh the transformative nature of that training and whether the model’s ability to produce vast quantities of competing works creates cognizable market harm.
On September 29, 2026, the Third Circuit affirmed Thomson Reuters’ partial summary judgment against ROSS Intelligence in the first federal appellate decision to address fair use in AI training. The case concerns Westlaw headnotes, short summaries written by Thomson Reuters editors to identify points of law in judicial opinions.
ROSS Intelligence developed an AI legal-search engine that answered plain-language legal questions by returning relevant passages from existing judicial opinions. To train its system, ROSS Intelligence hired a third party to prepare roughly 25,000 training memoranda. The memoranda used Westlaw headnotes to frame legal questions and paired those questions with judicial-opinion passages labeled according to how responsive they were. ROSS Intelligence then converted the memoranda into machine-readable data so its AI product could learn which opinion passages were responsive to particular legal questions. ROSS Intelligence intended the finished product to compete directly with Westlaw.
The central legal question was whether ROSS Intelligence’s copying of Westlaw headnotes qualified as fair use. Copyright law directs courts to consider four factors: (1) the purpose and character of the use, including whether it is commercial and transformative; (2) the nature of the copyrighted work; (3) the amount of the copyrighted work used and whether that amount was justified; and (4) the effect of the use on existing or reasonably expected markets for the copyrighted work.
Before diving into the fair use analysis, the Court answered the threshold question of whether the Westlaw headnotes were sufficiently original to warrant copyright protection in the affirmative. Because the headnotes had “some creative spark,” such as reflecting decisions about which points should be included and how to work those points, they met the admittedly low bar for originality under copyright law. This decision reinforces an important concept that even works that may seem unoriginal or “factual” in nature at first glance can still qualify for copyright protection.
The District of Delaware and the Third Circuit agreed that ROSS Intelligence’s copying did not qualify as fair use, with the analysis turning primarily on the first and fourth factors.
The First Factor Weighed Against Fair Use Because ROSS Intelligence’s Use Was Only Minimally Transformative
The first factor focuses on the purpose and character of the challenged use, including whether the use is commercial and whether it serves a new, transformative purpose. The Third Circuit held that this factor weighed against fair use because ROSS Intelligence’s use of the Westlaw headnotes was highly commercial and only minimally transformative.
Westlaw uses headnotes to help lawyers identify relevant law. ROSS Intelligence used those same headnotes to train its AI product to identify relevant judicial passages. Although the headnotes appeared only in the training process, the Third Circuit concluded that ROSS Intelligence’s use of the headnotes and Thomson Reuters’ use of the headnotes served essentially the same purpose: improving a legal-research service that helps lawyers find relevant law. The intermediate training step therefore did not materially change the purpose or character of the use. The court also rejected ROSS Intelligence’s analogies to cases discussing “intermediate copying” in the context of computer code to make software operable, finding that it was not actually necessary for ROSS Intelligence to copy the headnotes to access any underlying unprotected information (the legal opinions). Copying the headnotes was the “easy” way to create the training memos, but it was not necessary.
The Second and Third Factors Split and Played a Limited Role
The second and third factors pointed in opposite directions and played a limited role in the outcome, as is often the case in a fair use analysis. The second factor considers the nature of the copyrighted work. The court held that this factor slightly favored ROSS Intelligence because Westlaw headnotes summarize judicial opinions and legal rules, which is “more factual.” The third factor considers how much of the copyrighted work was copied and whether that amount was justified. Because ROSS Intelligence copied entire headnotes, the Third Circuit held that the third factor favored Thomson Reuters.
The Fourth Factor Weighed Against Fair Use Because ROSS Intelligence’s Conduct Harmed Thomson Reuters in Two Specific Markets
The fourth factor focuses on whether the challenged conduct harms an existing or reasonably expected market for the copyrighted work. As part of this analysis, the Third Circuit identified four separate factors: harm to the original market, harm to the value of the copyrighted work, harm to the potential derivative market, and the alleged public benefits of the copying. The Third Circuit held that this factor weighed against fair use because ROSS Intelligence’s conduct threatened two cognizable markets.
First, the court identified harm to Thomson Reuters in the existing legal-research market. ROSS Intelligence used Westlaw headnotes to build an AI legal-research product designed to compete directly with Westlaw. The court reasoned that widespread use of the headnotes to build substitute products would reduce the value of Thomson Reuters’ editorial work in that market.
Second, the court identified harm to Thomson Reuters in a developing market for licensing headnotes as AI-training data. Thomson Reuters already used headnotes to train its own AI search products, and the court concluded that ROSS Intelligence’s unlicensed use deprived Thomson Reuters of the opportunity to license the headnotes for the same purpose. Notably, the court acknowledged that the market for licensing headnotes to train AI is “rapidly developing.”
Having identified harm in both markets, the Third Circuit concluded that the fourth factor weighed against fair use.
With the first, third, and fourth factors weighing against fair use—and only the second factor slightly favoring ROSS Intelligence—the Third Circuit affirmed the district court’s ruling that ROSS Intelligence’s copying did not qualify as fair use.
The Core Copyright Questions for Generative AI Remain Unresolved
The Third Circuit led its opinion by claiming that “this is no more than an ordinary copyright case,” but simultaneously recognized that “[a]rtificial intelligence (‘AI’) is a powerful machine learning technology that is poised to impact many facets of American life.” ROSS Intelligence gave the Third Circuit its first opportunity to address fair use in AI training, but not the questions at the center of generative AI litigation. ROSS Intelligence’s AI product returned existing judicial passages; it did not learn from copyrighted works to generate new expression. The court therefore was never presented with the central generative AI questions: Is training a model on copyrighted works transformative because it resembles human learning? And if the model can then produce vast quantities of competing content, is that ordinary competition copyright law should permit or a cognizable market harm? In its analysis of the first fair use factor, the Third Circuit explicitly said that the concerns raised in a separate generative AI case do not apply to ROSS Intelligence. Nevertheless, the opinion suggests that the differences may be influential for the first factor: “Unlike the AI models in Bartz and In re: OpenAI, ROSS’s AI platform cannot generate original expression, and the evidence here supports the opposite conclusion about transformativeness.”
As for the fourth fair use factor, district courts have already divided over the market-harm question. In Bartz v. Anthropic PBC, Judge William Alsup compared training Claude on books to teaching schoolchildren to write: copyright protects authors against copying, not against competition from new writers who learned from existing works. In Kadrey v. Meta Platforms, Inc., Judge Vince Chhabria rejected that analogy as incomplete, reasoning that an LLM can generate “countless competing works” at a speed and scale no human student can match. No federal appellate court has resolved that disagreement.
The broader question remains for a future appellate court: should copyright law treat generative AI training more like human learning, or does the model’s capacity to generate competing expression at unprecedented scale require different analyses for purpose of use and market harm?
Related Insights
Subscribe to Ballard Spahr Mailing Lists
Get the latest significant legal alerts, news, webinars, and insights that affect your industry.
Subscribe Copyright © 2026 by Ballard Spahr LLP.
www.ballardspahr.com
(No claim to original U.S. government material.)
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, including electronic, mechanical, photocopying, recording, or otherwise, without prior written permission of the author and publisher.
This alert is a periodic publication of Ballard Spahr LLP and is intended to notify recipients of new developments in the law. It should not be construed as legal advice or legal opinion on any specific facts or circumstances. The contents are intended for general informational purposes only, and you are urged to consult your own attorney concerning your situation and specific legal questions you have.