Yesterday, in the first U.S. ruling on the closely scrutinized question of fair use in the AI-related copyright litigation context, U.S. Circuit Judge Stephanos Bilbas, sitting in the U.S. District Court for the District of Delaware by designation from the Third Circuit, held that U.S. copyright law did not permit Thomson Reuters’s now-defunct former competitor, Ross Intelligence, to copy Thomson Reuters’s content to build a competing artificial intelligence-based platform.
The fair use doctrine in copyright law allows limited, unauthorized use of copyrighted material in certain circumstances. The doctrine is an affirmative defense frequently invoked by technology companies arguing that their generative AI systems make limited fair use of third parties’ copyrighted material by examining it to learn to create new content.
Thomson Reuters, the owner of the Westlaw legal research platform, sued Ross, a new Westlaw competitor that had developed a legal-research engine using artificial intelligence, for copyright infringement. To train its AI search tool, Ross required a database of legal questions and answers and sought to license Westlaw’s content. Considering Ross’s position as a competitor, Thomson Reuters denied Ross’s request. Ross, however, nonetheless built its competing product using Bulk Memos, which were, in turn, built from Westlaw headnotes.
In 2023, Judge Bilbas denied Thomson Reuters’s initial motion for summary judgment on its claim of copyright infringement and did not reject Ross’s fair use defense. That ruling was based, in large part, on Judge Bilbas’s belief that the analysis of the fourth factor of the fair use defense, which addresses the issue of market impact, should be left for the jury.
Judge Bilbas thought “Ross’s use might be transformative, creating a brand-new search platform that serves a different purpose than Westlaw.” Thomson Reuters Enter. Ctr. GmbH, et al. v. Ross Intelligence, Inc., Case No. 1:20-cv-613-SB (D. Del. Feb. 11, 2025), at 22. He reasoned that, if true, “then Ross would not be a market substitute for Westlaw.” Id. Judge Bilbas also cited concerns about “whether there was a relevant, genuine issue of material fact about whether Thomson Reuters would use its data to train AI tools or sell its headnotes as training data.” Id.
In reversing his 2023 ruling, Judge Bilbas noted that, in preparing for the August 2024 trial date, he studied the case materials more closely and “realized that [his] prior summary-judgment ruling had not gone far enough.” Id. at 3. Thus, he invited the parties to renew their summary judgment briefing.
This time around, Judge Bilbas rejected all Ross’s defenses, including fair use, innocent infringement, and copyright misuse. In so ruling, he stated that, in hindsight, his prior concerns were unpersuasive: “Even taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute. 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.” Id. at 22 (emphasis added). Notably, Judge Bilbas cited the fact that Ross bore the burden of proof and had not proved “enough facts to show that these markets do not exist and would not be affected.” Id.
Judge Bilas further rejected Ross’s argument that the public benefit that copying Westlaw data may produce warranted application of the fair use defense: “Yes, there is a public interest in accessing the law. But legal opinions are freely available, and ‘the public’s interest in the subject matter’ alone is not enough…. The public has no right to Thomson Reuters’s parsing of the law.” Id. at 22-23.
While the technology at issue in this case did not involve generative AI technology, which Judge Bilbas specifically noted, this decision is the first to address whether training an AI model on copyrighted content is fair use. As such, Judge Bilbas’s findings could prove instructive in other cases involving generative AI technology.