Allegro

A.I. LEGAL UPDATE

Volume 125, No. 3March, 2025

Harvey S. Mars

This month’s submission for Local 802’s A.I. series is by Local 802 Recording Vice President Harvey Mars, the officer representative to Local 802’s A.I. Committee.

In one of the first rulings concerning the legality of using copyrighted data to train artificial intelligence models, Third Circuit Judge Stephanos Bibas on February 11, 2025 rejected the “fair use” doctrine as a defense to the utilization of copyrighted material in creation of an A.I database. The decision was issued in Thompson Reuters v. Ross Intelligence Inc., No. 1:20-cv-613-SB (U.S.D.C., Del 2025).

In 2020, Thomson Reuters had sued the now-defunct A.I. start-up Ross Intelligence (it went bankrupt as a result of litigation costs) for alleged improper use of Thomson Reuters materials, including case headnotes in its Westlaw search engine, to train its new AI model. Westlaw is a legal research database that catalogues judicial decisions through annotated case headnotes that detail and pinpoint the legal holdings in these decisions. Thomas Reuters had rejected Ross’ request to license the headnotes to create its own legal research database. When that request was denied, Ross hired a third party, Legalease, to cull together Westlaw headnotes and summarize them for training Ross’ new data base. Ross defended against the copyright infringement claim by arguing that it was entitled to use this material.

A key issue before the court was whether Ross Intelligence’s usage of headnotes constituted fair use, which permits a person to use portions of another’s work in limited circumstances without infringing on their copyright. Courts use a four-pronged test to determine whether a defendant can successfully use the fair use defense: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) how much of the work was copied and was that a substantial part of the entire work; and (4) whether the defendant’s use of the work affected its value.

In this case, Judge Bibas determined that each side had two of these factors in their favor. But the fourth factor, which supported Thomson Reuters, weighed most heavily in his finding that the fair use defense was inapplicable because Ross Intelligence sought to develop a competitive product. Clearly, Ross’ use of Westlaw’s material would have a profound impact on the value of the Westlaw search engine. This tipped the scales against a fair use defense.

It is noteworthy that Judge Bibas held that his earlier holding that consideration of the fourth factor needed to be decided by a jury was wrong. In overruling himself he stated in his ruling granting partial summary judgment that “a smart man known when he is right; a wise man knows when he is wrong. Wisdom does not always find me, so I try to embrace it when it does — even when it comes late, as it did here.” This is the first time I have ever encountered a judge acknowledging they had decided a matter wrongly.

Lawsuits against other companies, like OpenAI and Microsoft, are currently pending in courts throughout the country, and decisions in those cases may involve similar questions about the fair use defense. However, Judge Bibas noted that Ross Intelligence’s A.I. model was not actually a generative one and that his decision was based only on Ross’s non-generative AI model. The distinction between the training data and resulting outputs from generative and non-generative A.I. may likely be key to deciding future cases. We will be monitoring these cases to see how the fair use doctrine is applied there as well.

Send feedback on Local 802’s A.I. series to Allegro@Local802afm.org.

CLICK HERE TO JOIN OUR A.I. COMMITTEE E-MAIL LIST AND TO LEARN ABOUT UPCOMING MEETINGS