On Tuesday, in Thomson Reuters Enterprise Centre GmbH et al. v. ROSS Intelligence Inc., Judge Stephanos Bilbas (sitting in the Delaware Federal District Court by designation from the Third Circuit) granted partial summary judgment in favor of Plaintiff Thomson Reuters after renewed briefing, finding that ROSS’s use of certain Westlaw headnotes infringed Thompson Reuters’ copyrights as a matter of law, rejecting its affirmative defenses and concluding that such use was not a fair use under the Copyright Act. Judge Bilbas’ opinion is remarkable not only for its conclusive holdings, but because it represents a near-complete turnabout from the Court’s prior holding on the same questions two years ago, noting that “a wise man knows when he is wrong.” Thomson Reuters Enterprise Centre GmbH et al. v. ROSS Intelligence Inc., No. 1:20-cv-613, at 2 (D. Del. Feb. 11, 2025).

As the Opinion summarizes, in 2020, Thomson Reuters and West Publishing Corp. sued ROSS, alleging that ROSS copied a sizable amount of proprietary Westlaw content, including its unique key number system and headnotes found in their database, for development in ROSS’s own legal research tool. ROSS built its rival product, in part, by using “Bulk Memos” obtained from another company, LegalEase, which themselves were built from Westlaw headnotes.
 
The Opinion concluded that Thompson Reuters’ copyrights in its headnotes were valid, and that a subset comprising 2,243 headnotes was actually copied as a matter of law.  In doing so, the Opinion summarily dismissed ROSS’s claims of innocent infringement (noting that innocent infringement has no bearing on liability, only damages, and that the defense is further obviated by including a copyright notice), copyright misuse, merger, and scenes a faire. From there, the Court considered each of the statutory fair use factors and determined that ROSS’s use of Westlaw headnotes was not fair. The fair use factors, set forth in the Copyright Act and considered by courts as a balancing test, are: (1) the purpose and character of the work (e.g., commercial vs nonprofit), (2) the nature of the work, (3) the amount and substantiality of the work taken relative to the work’s whole, and (4) the use’s potential effects on the work’s value or the market. 17 U.S.C. § 107.  
 
Here, the Court found that factors (2) and (3) favored ROSS, but on balance, rejected ROSS’s claim of fair use based largely on first and fourth factors. As for the first factor, the Court found that ROSS’s use was commercial and non-transformative. Judge Bilbas rejected ROSS’s reliance on “intermediate copying,” case law, noting that these cases depend upon the notion that the copying is necessary to reach the underlying ideas, as opposed to ROSS’s use of Westlaw content to merely make the development of its competing platform easier. The fourth factor, generally viewed to be the most significant, was also found to weigh in favor of infringement and against fair use. The Court identified the relevant markets for the copyrighted works as legal research platforms (the primary market) and data to train legal AI models (a potential derivative market), and determined that the unauthorized use of Westlaw’s content impacted (or would impact) both. While Thomson Reuters has used the same data to train their own search tools, ROSS developed their tool as a market substitute to Westlaw, and thus “the effect on a potential market for AI training data is enough.”
 
Takeaways
The Court was careful to emphasize that this case does not involve generative AI; however, its impact may prove significant for all uses of third-party content to train AI. Those who develop, train, and use AI models should continue to exercise caution with respect to where and how training data is sourced, including through investigation and suitable contractual protections.
 
Baker Botts will continue to track developments and provide timely updates and guidance.
 
*Benjamin Bafumi, a law clerk at Baker Botts, assisted in the preparation of this article.