Appeals Court Affirms Thomson Reuters AI Copyright Victory

Overview

A United States federal appeals court has upheld a landmark copyright victory for Thomson Reuters, dismissing the final legal challenges from the now-defunct legal AI startup Ross Intelligence. The ruling, issued by the 3rd U.S. Circuit Court of Appeals, confirms that using copyrighted legal materials to train a commercial competing product without permission or compensation constitutes infringement.

This decision is widely regarded as the first federal appeals-level judgment to reject a fair use defense in the context of AI training data. It establishes a significant legal precedent for content publishers seeking to protect their intellectual property against companies building competitive AI tools.

“The undisputed evidence reflects that Thomson Reuters’s materials possess a creative spark, and ROSS aspired to be a direct competitor by using them for a highly similar purpose.”

— Judge Tamika Montgomery-Reeves, 3rd U.S. Circuit Court of Appeals

The Core Legal Dispute

The lawsuit originated in 2020, preceding the public launch of generative chatbots like ChatGPT. Thomson Reuters alleged that Ross Intelligence had copied thousands of “headnotes”—brief summaries of legal issues in judicial opinions—from its Westlaw platform. These headnotes were used to train Ross’s proprietary AI-powered legal search engine.

Unlike modern generative AI models that create original text, Ross’s system functioned as a retrieval-based search engine. It pulled existing passages from court documents to answer user queries. Despite this technological distinction, the court found that the fundamental economic harm to Thomson Reuters remained identical to that of any other copyright infringement.

Judge Stephanos Bibas, who presided over the initial district court ruling in February 2025, emphasized the competitive nature of Ross’s actions. He noted that even when viewing facts in favor of the defendant, Ross intended to create a market substitute for Westlaw.

“And it does not matter whether Thomson Reuters has used the data to train its own legal search tools; the effect on a potential market for AI training data is enough.”

— Judge Stephanos Bibas

The appeals court agreed with this assessment, reinforcing the principle that the commercial intent and market substitution effects are critical factors in fair use analysis.

Distinction from Generative AI Litigation

While the ruling is a major win for publishers, legal experts note that its scope may be limited. Judge Montgomery-Reeves explicitly clarified that this case does not involve a large-language model (LLM) or a generative AI tool that produces new, original expression.

Ross’s AI did not rewrite or generate novel summaries; it retrieved and reproduced existing text. The judge argued that this distinction may narrow the ruling’s applicability to the broader wave of copyright lawsuits currently pending against companies like OpenAI and Anthropic.

In previous cases, such as Bartz v. Anthropic, judges have found that training LLMs on published works can constitute fair use, particularly when the output is transformative. The $1.5 billion settlement in that case was primarily tied to Anthropic’s decision to store pirated copies of books, rather than the act of training itself.

“Unlike the AI models in Bartz… ROSS’s AI platform cannot generate original expression, and the evidence here supports the opposite conclusion about transformativeness.”

— Judge Tamika Montgomery-Reeves

This language suggests that if Ross had developed a system capable of generating original summaries with its own creative expression, the outcome might have differed. The ruling effectively draws a line between retrieval-based AI and generative AI regarding fair use considerations.

Implications for the Media and Publishing Industry

The decision has been closely watched by news publishers and legal scholars as a potential blueprint for future litigation. While it may not directly resolve cases involving generative AI, it strengthens the legal footing for publishers arguing that unauthorized scraping and use of their content for competitive purposes is unlawful.

Key takeaways for the industry include:

• Market substitution is pivotal: Courts are willing to consider the impact on potential markets for licensed data, not just the current market for the original work.
• Creative effort matters: The court acknowledged the “creative spark” in editorial summaries, distinguishing them from raw factual data that may not be copyrightable.
• Fair use is not absolute: The ruling rejects the notion that using copyrighted material to build a competing commercial product is automatically permissible under fair use doctrines.

Although Ross Intelligence is no longer operational, the legal principles affirmed in this appeal remain relevant. They provide ammunition for ongoing disputes between major tech firms and media outlets, where the boundary between innovation and infringement continues to be tested in federal courts.