A federal appeals court has issued the first U.S. appellate ruling on fair use in an artificial-intelligence context, rejecting a legal research company’s attempt to use copyrighted Westlaw material to build a competing AI-based search product. The 3rd U.S. Circuit Court of Appeals unanimously ruled that Ross Intelligence’s use of Thomson Reuters headnotes was not transformative and therefore did not qualify as fair use.
The decision is significant because technology companies are relying heavily on the fair-use doctrine in a growing wave of lawsuits over AI training. The appeals court nevertheless stressed that the Ross case differs from disputes involving generative AI systems such as ChatGPT, limiting how far the ruling can be applied to the broader industry.
AI Fair Use Ruling Rejects Ross Intelligence Defense
Ross Intelligence used Westlaw headnotes, which are editorial summaries of legal points contained in court opinions, while developing a competing legal research service. Thomson Reuters sued in 2020, arguing that Ross had copied thousands of those summaries without permission to train its system.
The 3rd Circuit agreed with a Delaware federal court that the headnotes were sufficiently original to qualify for copyright protection. It also concluded that Ross used them for a purpose too similar to Thomson Reuters’ own commercial use, weakening Ross’s argument that its copying transformed the material into something fundamentally different.
Judge Tamika Montgomery-Reeves wrote for the unanimous panel that the dispute was ultimately an ordinary copyright case despite its association with AI technology. The court focused on the nature of the copied material, Ross’s purpose and the fact that Ross aimed to compete directly in the legal research market.
The Decision Is the First Appellate AI Fair Use Case
The ruling matters because no federal appeals court had previously addressed fair use in an AI-related training dispute. Trial courts have issued several important decisions, but appellate precedent carries greater weight because it can guide lower courts across an entire federal circuit.
Fair use allows copyrighted material to be used without permission in some circumstances, depending on factors including purpose, transformation, amount copied and market impact. AI companies frequently argue that training models on large datasets is transformative because the resulting systems do not simply reproduce the original works but learn statistical patterns that support new outputs.
The 3rd Circuit did not accept that reasoning for Ross because the company’s system was built to compete with Westlaw in legal research. The court concluded that Ross’s use threatened a market closely related to the one served by the copyrighted headnotes, which weighed heavily against fair use.
The Court Drew a Line Around Generative AI
The most important limitation is that Ross’s technology was not generative AI. The system did not operate like modern large language models that create new text, images or other outputs from broad training datasets.
The appeals court explicitly distinguished the case from lawsuits involving generative AI and noted that concerns raised by the Justice Department in litigation against OpenAI did not apply in the same way. That distinction means the ruling is not a definitive answer to whether companies such as OpenAI, Anthropic, Meta or Google may rely on fair use when training generative models.
The court’s caution is likely to be central in future litigation. Copyright owners will point to the ruling as evidence that commercial AI training can fail fair-use analysis, while AI companies will emphasize that Ross involved a direct competitor using editorial content for a closely similar purpose.
Copyright Owners Gain an Important Precedent
The decision gives publishers, authors and other rights holders a new appellate precedent supporting the argument that AI developers cannot assume training is automatically transformative. When copied material is used to build a competing product serving the same market, courts may be more willing to find infringement.
Thomson Reuters said the ruling confirmed that Westlaw’s attorney-created editorial content is protected by copyright. The company argued that strong copyright protection supports innovation by preserving incentives to create and maintain high-value professional information products.
Ross shut down its platform in 2021, citing the cost of the litigation. The case nevertheless continued because of the larger legal questions and the potential damages associated with the alleged copying.
Generative AI Companies Still Face Separate Lawsuits
OpenAI, Microsoft, Meta and other technology companies are defending separate copyright cases brought by publishers, authors, visual artists and other creators. Those disputes often involve much larger training datasets and systems capable of generating new content rather than simply locating or organizing information.
Courts in those cases will have to decide whether using millions of copyrighted works to train general-purpose models is sufficiently transformative and whether resulting products substitute for the original material in relevant markets. Those questions remain unsettled despite the 3rd Circuit’s new ruling.
The issue has become economically important because licensing every work in a large AI training dataset could be expensive or technically difficult, while creators argue that unrestricted copying lets technology companies build valuable products without paying for the underlying content.
The Market Effect Could Become the Decisive Factor
The Ross ruling highlights the importance of market harm in fair-use analysis. Courts are more skeptical when a new product directly competes with the source of the copied material or reduces demand for licensed access to that content.
That could become a major battleground in generative AI cases. Publishers argue that AI systems can substitute for news, books, images and professional research, while technology companies contend that their models create different products and often increase the usefulness of underlying information.
Evidence about customer behavior, licensing markets and whether AI outputs replace original works may therefore become as important as technical arguments about how models are trained. The legal question is not merely whether an AI system copies data during training but what commercial purpose that copying serves and what market it affects.
The Precedent Is Important but Narrow
The AI fair use ruling gives courts and copyright owners a meaningful appellate precedent, but it does not settle the wider debate over generative AI training. The 3rd Circuit deliberately limited its reasoning to a system that used copyrighted legal summaries to build a directly competing research tool.
That narrowness may make the decision more influential rather than less, because future courts can use its framework without pretending that every AI system works the same way. Companies will need to show not only that their technology is novel but also that their use of copyrighted material is meaningfully transformative and does not improperly damage protected markets.
The result is a clearer warning for AI developers but not a universal rule. Fair use remains a fact-specific doctrine, and the next major cases involving generative AI will determine how broadly this first appellate precedent ultimately reaches.
