Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc.

In case involving creation of materials for AI training, Third Circuit affirms summary judgment in favor of Thomson Reuters (creator of Westlaw) on protectability of Westlaw’s “headnotes” given each headnote had enough “creative spark” as product of editors’ creative decisions and rejects fair use defense where allegedly infringing works (memos created from Westlaw headnotes) were to be used for same purpose as headnotes and harmed potential market for Westlaw and AI training materials could have been created without copying. 

Thomson Reuters Enterprise Centre GmbH is the creator of the legal research platform Westlaw and holds a copyright in its content, including the headnotes that appear at the beginning of opinions. These headnotes, which each reflect a point of law from the judicial opinion, are prepared by editors based on specific guidelines. ROSS is a startup legal research platform that sought to create a search engine capable of providing relevant passages of legal opinions in response to plain-language questions. To do that, ROSS uses AI, which had to be trained on what made a judicial opinion responsive to legal questions. ROSS hired another company to create thousands of legal memos that set out questions along with responsive passages from court opinions. To create these legal memos, the company used thousands of Westlaw headnotes.

After learning about ROSS’ actions, Thomson Reuters sued for, among other claims, copyright infringement. After initially denying the parties’ cross-motions for summary judgment in substantial part, holding that there were factual issues as to whether the headnotes were sufficiently original to warrant copyright protection, as well as factual issues on various elements of ROSS’ fair use defense (read our summary of the court’s earlier decision here), the district court reconsidered its earlier summary judgment decision sua sponte and granted partial summary judgment in Thomson Reuters’ favor, holding that ROSS had infringed on thousands of Westlaw headnotes and that its doing so did not constitute fair use. (Read our summary of the court’s decision here.) The district court then certified two questions for interlocutory review by the Third Circuit: (1) whether the headnotes are original (a prerequisite for copyright protection) and (2) whether ROSS’ use of the headnotes is protected by the fair use doctrine.

As an initial matter, the Third Circuit explained that this case was an ordinary copyright case, distinguishable from various cases in other circuits about generative AI and large language models (LLMs). With that framing in mind, the court affirmed. First, as to originality, the court explained that the Westlaw headnotes at issue in the appeal had sufficient “creative spark” to qualify for copyright protection. Each headnote was created through various editorial decisions, such as which points of law are important enough to get a headnote and how each point of law should be conveyed. In finding sufficient originality, the court rejected ROSS’ arguments that doing so would give Westlaw a monopoly on the law because, despite ROSS’ arguments, judicial opinions remain free for publication by anyone. It also found that the merger doctrine, which precludes copyright protection where a work’s underlying idea can only be expressed in one way or a few ways, did not apply as there are many ways to express points of law from judicial opinions.

The court then turned to fair use, analyzing each of the four recognized factors. First, as to the purpose and character of the use, the court found that not only was ROSS’ use commercial (indeed, it sought to create a competitor to Westlaw), but it was, at best, minimally transformative given that its use of the headnotes shared the same ultimate purpose as Thomson Reuters’ (creating and optimizing a legal research platform). That ROSS’ use involved the intermediate step of training an AI search engine did not make its use transformative unlike, perhaps, where materials are used to train AI LLMs that can generate original responses. The court also rejected ROSS’ arguments comparing its copying to the use of computer code to make compatible systems because ROSS’ copying was, unlike in those cases, unnecessary—it could have created its memos from scratch.

On the second factor, which examines the nature of the copyrighted work, the court found for ROSS. The Westlaw headnotes, the court explained, are more fact than fiction and, as such, fall further from the core purpose of copyright protection. However, the third factor, which looks at the amount and substantiality of the portion of the copyrighted material used and asks whether that was reasonable in relation to the purpose of the copying, struck in favor of Thompson Reuters where ROSS’ use was entirely unnecessary. By definition, ROSS took more from Westlaw than was necessary, as it could have simply created its legal memos from publicly available judicial opinions.

Finally, on the fourth factor—the effect of the use on the potential market for or value of the copyrighted work—the court found that ROSS’ use harmed the market of the headnotes in multiple ways without any public benefits. On the original market, ROSS sought to create a competitor for Westlaw, and by creating a competing legal research platform, ROSS used the headnotes in a way that diminished the value of the originals that draw users onto the Westlaw platform. ROSS’ use also harmed the potential derivative market to license the headnotes as an AI training tool. And the public was not being served where ROSS was offering its service at prices comparable to Westlaw’s, and there was no evidence that the district court decision would hamper AI development or harm national security, despite ROSS’ arguments.

With the first, third and fourth factors weighing against fair use, the court held that ROSS’ use was not fair.