Thomson Reuters v. Ross Intelligence

A conflict between innovation and copyright protection

The case dates back to 2020, when Ross Intelligence, a startup specializing in legal research using artificial intelligence, was accused of using case law summaries from Westlaw, a platform owned by Thomson Reuters, without authorization. These summaries, known as headnotes, are written by experts and structure court decisions to facilitate research. Ross Intelligence is said to have copied this content to train its AI and improve the relevance of its legal search engine.

Judge Stephanos Bibas ruled that these summaries, although based on decisions in the public domain, enjoyed specific protection because of the human work involved in selecting and writing them. He rejected Ross’s argument that these summaries constituted mere “added noise” in AI training, pointing out that editorial work was carried out to synthesize and organize the information, which was sufficient to justify copyright protection.

Rejection of fair use: a major decision for content protection

On the question of fair use, the judge reminds us that we must take into account :

• The purpose and nature of the use, in particular whether it is commercial or non-profit: the judge considers Ross’s use to be commercial.

On this point, the court concluded that Ross Intelligence’s use was commercial and not transformative, due to the creation of a direct competitor to Thomson Reuters.

• The nature of the protected work

On this point, the judge considers that the abstracts do contain creative elements, but they are far from being among the most creative works, so this second point is favorable to Ross.

• The quantity of the work used and the importance of the part used in relation to the protected work as a whole

On this point, the judge ruled in favor of the infringer. Indeed, his final production does not include a summary of Thomson’s decisions. Since Ross did not make the summaries available to the public, this third point is in his favor.

• The impact of Ross’ use on the value or potential market for the copyrighted work. 17 U.S.C. §107(1)-(4)

On this last point, the judge held that by exploiting Westlaw’s content to develop a competing product, Ross Intelligence was undermining Thomson Reuters’ business model.

Judge Bibas also distinguished this case from precedents on software copying, where duplication of code was considered necessary to access protected functional elements. Here, the Court emphasized that the reproduction of case law summaries was not essential to access the underlying ideas of judicial decisions. Moreover, whereas software pursues a purely functional objective, this case concerns editorial and literary choices, which are protected by copyright.

Thus, although the other two factors could be discussed, the judge concluded that fair use was legally inapplicable, the first and fourth factors being decisive and unfavorable to Ross Intelligence.

A global challenge for AI management in the United States

Issued on the same day as the closing of the World Summit on AI, the ruling marks a victory for copyright holders, even though the US refused to sign the summit’s final convention, arguing for a light-handed approach to regulation to avoid holding back a booming industry.

This context contrasts with Europe, where initiatives such as the AI Act seek to provide a stricter framework for the use of protected content in training AI models.

This decision could have a major impact on the development of AI, while other companies such as OpenAI and Microsoft are facing similar lawsuits for using protected content[2].

---

[1] Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc, Docket No. 1:20-cv-00613

[2] US District Court Southern District of New York, The New York Times Company v. Microsoft Corp, OpenAI Inc. et al, Dec. 27, 2023