Sculpting A Copyright Do-Over

It is the rare judicial opinion, particularly in the IP arena, that starts with the court admitting that they got an earlier decision wrong. Penning the opinion? None other than the Third Circuit’s Hon. Stephanos Bibas, sitting by designation in the District of Delaware. Widely (and rightly) hailed as one of the judiciary’s most incisive and eloquent opinion authors, Bibas’s February 11, 2025 opinion in Thomson Reuters v. Ross Intelligence does not disappoint. Not only does it provide a rare glimpse into the thought process of a judge unafraid to revisit and revise a prior opinion, it does so in a case involving material very familiar to the legal community. As a bonus, the opinion answers some pressing questions on one of the hottest topics in copyright law today. 

As at least some readers will realize, the plaintiff in the case is the owner of Westlaw, one of the leading legal research platforms. As part of its offerings, Westlaw takes published judicial opinions and organizes those opinions and its “editorial content and annotations,” (e.g., headnotes, under its “Key Number System, a numerical taxonomy”). Nothing new here to any lawyer. For its part, the defendant Ross tried to license Westlaw’s content to train its AI-search tool for its new legal research search engine. But Westlaw refused, leading Ross to train its AI using “Bulk Memos” that it purchased from a company called LegalEase. How were the Bulk Memos created? By having lawyers create questions based on Westlaw headnotes! Unsurprisingly, Thomson Reuters was not pleased and filed against Ross for copyright infringement.

After some years of litigation, Thomson Reuters moved for summary judgment on two grounds: copyright infringement and that Ross’s fair-use defense failed as a matter of law. That original motion was denied in large part and the case progressed toward a trial date last summer. As the court prepared for trial, however, it “studied the case materials more closely and realized that my prior summary-judgment ruling had not gone far enough. So I continued the trial and invited the parties to renew their summary judgment briefing.” Based on that renewed briefing, as well as the court’s painstaking review of “the Bulk Memo questions, headnotes, and opinions side by side,” a different result was reached. 

Concerning copyright infringement, the Court had previously found that “a jury would need to decide whether the headnotes and Key Number System were original enough” to qualify as copyrighted material. Upon further reflection of the seminal Supreme Court decision in Feist, however, the Court decided that there was “no genuine dispute that the headnotes and Key Number System clear Feist’s minimal threshold for originality.” First, the Court found that the headnotes represented an original compilation, but even further each headnote qualified as an “individual, copyrightable work.” To get there, the Court analogized uncopyrightable judicial opinions as a block of raw marble from which the headnote sculptor, i.e., Westlaw, could exercise “editorial expression” and create copyrightable works of value. Second, the Court applied similar reasoning to find that Westlaw’s Key Number System met the originality standard as a matter of law. With those reconsidered positions in tow, the Court then decided whether a batch of 2,830 headnotes identified by Thomson Reuters infringed. 

To start, the Court pointed to a concession from Ross’s expert that the “Bulk Memo questions for this batch closely resemble the headnotes’ text and that the headnotes differ significantly from the text of the judicial opinions” for a large percentage of the batch now under review for evidence of actual copying. Then, the court “compared how similar each of the 2,830 Bulk Memo questions, headnotes, and judicial opinions are, one by one” to see whether there was “strong circumstantial evidence of actual copying” by LegalEase when they created the Bulk Memos later sold to Ross. After an effort of “slogging through all 2,830 headnotes” the court found on summary judgment that there was “actual copying of 2,243” headnotes in the Bulk Memos. With that finding in hand, it was simple for the court to also grant summary judgment on substantial similarity for that same batch of copied headnotes, or those “headnotes whose language very closely tracks the language of the Bulk Memo question but not the language of the case opinion.” Left for trial were all the other disputed headnotes, as well as the determination of which of the 2,243 headnotes deemed infringed represented unexpired copyrights. 

Making matters worse for Ross, the court made quick work of various Ross defenses, such as innocent infringement and copyright misuse. The court also reconsidered its prior 2023 opinion on the question of fair use, based on “new information and understanding.” Finding first that “fair use is a question for the judge, not the jury,” the court found that Thomson Reuters had the better argument on the two most important of the four fair use factors, “and on the overall balancing.” On factor one, the court found that Ross’s use of the copyrighted material was commercial use and not transformative, “because it does not have a ‘further purpose or different character’ from Thomson Reuters’s.” Relevant to that finding was that Ross’s AI is not generative — creating new content — but just “spits back relevant judicial opinions that have already been written,” just like Westlaw’s headnotes and key number system is used for. Still, the fact that Ross’s use of the copyrighted material did “not appear as part of the final product that Ross put forward to consumers” merited further attention. But after distinguishing Ross’s cited cases on “intermediate copying” in the software code context, the court found no fair use because Ross’s use of the copyrighted material was not necessary to launch its research product. Accordingly, the use by Ross was not transformative and no longer a question for the jury as the court had previously concluded.

Similarly, the court found that it had been wrong to leave for the jury the question of whether Ross had created “a brand-new research platform that serves a different purpose than Westlaw.” In relevant part, the court found that “[e]ven taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute.” While the public has an interest in accessing the law, here, legal opinions were already “freely available.” Because there was “nothing that Thomson Reuters created that Ross could not have created for itself or hired LegalEase to create for it without infringing Thomson Reuters’s copyrights,” factor four also went against Ross. And since both of the more important factors, one and four, favored Thomson Reuters, the court entered summary judgment against Ross’s fair use defense.

Ultimately, this decision has relevance for the wide swath of courts considering use of copyrighted material to train generative AI models, even as Bibas made clear that he was not considering a generative AI model in dealing with the case before him. We are waiting to see how the current crop of cases dealing with those questions will come out. For now, at least, we have some indication of how a wise judge would begin to arrive at the answers.

Please feel free to send comments or questions to me at gkroub@kskiplaw.com or via Twitter: @gkroub. Any topic suggestions or thoughts are most welcome.

Gaston Kroub lives in Brooklyn and is a founding partner of Kroub, Silbersher & Kolmykov PLLC, an intellectual property litigation boutique, and Markman Advisors LLC, a leading consultancy on patent issues for the investment community. Gaston’s practice focuses on intellectual property litigation and related counseling, with a strong focus on patent matters. You can reach him at gkroub@kskiplaw.com or follow him on Twitter: @gkroub.

The post Sculpting A Copyright Do-Over appeared first on Above the Law.