Thomson Reuters v. Ross Intelligence case

On Tuesday, February 11, 2025, the U.S. District Court for of Delaware issued a significant decision on copyright law and artificial intelligence. The litigation opposed Thomson Reuters, owner of a legal research platform and Ross Intelligence, a legal AI company. The Court ruled that Ross’ use of Westlaw’s headnotes to train its legal AI constituted copyright infringement and did not fall within the scope of fair use. 

 A conflict between innovation and copyright protection

The dispute arose when Ross intelligence trained its AI-powered legal research engine using and copying Thomson Reuters’ Westlaw database headnotes. These are concise summaries of court decisions from the public domain, written and summarized by human experts.

According to the judge Bibas, the headnotes can introduce creativity by distilling, synthesizing, or explaining part of an opinion. Furthermore, they are an original compilation using a minimal degree of creativity because of Thomson Reuters’s selection and arrangement.

Rejection of fair use defense: a landmark decision for copyright protection  

On the issue of fair use, the judge must consider four factors:

• The use’s purpose and character, including whether it is commercial or nonprofit.

On this point, the court concluded that Ross Intelligence’s use was commercial and not transformative. Indeed, Ross was using Thomson Reuters’ headnotes as AI data to create a legal research tool directly competing with Reuters platform. Therefore, factor one goes to Reuters.

• The copyrighted work’s nature.

On this point, the court concluded that Westlaw’s material has more than the minimal spark of originality required for copyright validity, “but the material is not that creative”. Therefore, factor two goes for Ross.

• How much of the work was used and how substantial a part it was relative to the copyrighted work’s whole.

On this point again, the judge sides in favor of the infringer. Indeed, Ross did not make Westlaw headnotes available to the public in its final product, thus benefiting from factor three.

• How Ross’s use affected the copyrighted work’s value or potential market.

The judge recalls that factor four “is undoubtedly the single most important element of fair use.” On this last point, the judge outlined that Ross’ use was not transformative, because it did not create a brand-new research platform. For that reason, Ross constituted a market substitute for Westlaw, and consequently negatively affected its business model.

In addition, the Judge differentiated this case from the others concerning software’s copies where copying was necessary for competitors to innovate. In Reuters’ case, the court outlined that copying was not reasonably necessary to achieve the user’s purpose. Indeed, there is no computer code where the underlying ideas cannot be reached without copying their expression.

Weighing all the factors together, the judge granted that factor four matters more than the others. Because it favored Thomson Reuters, the judge granted summary judgment for Thomson Reuters on fair use.

A global challenge for AI regulation in the US

Issued on the same day as the closing of the Paris AI Action Summit, this decision marks a victory for copyright holders, even though the United States refused to sign the summit’s final declaration, arguing for a light-handed approach to regulation to avoid holding back a booming industry.

This stance contrasts with Europe, where regulatory initiatives like the AI Act aim to impose stricter oversight on the use of copyrighted content in AI training.

This decision could have a significant impact on AI development, while other companies such as OpenAI and Microsoft are facing similar lawsuits over their use of copyrighted materials to train AI models.