A federal judge in Delaware has sided with Thomson Reuters in an artificial intelligence (“AI”)-centric copyright case, finding that a competing legal research site infringed Reuters’ copyright protected materials by using them to train its AI model. In addition to granting Reuters’ bid for partial summary judgment on its copyright infringement claim, the court found that the-now-defunct ROSS Intelligence’s copying of Reuters’ case law headnotes to train its large language model is not fair use. In siding with Reuters, Judge Stephanos Bibas walked back on his 2023 summary judgment opinion, in which he held that infringement and fair use were issues for the jury.
A Bit of Background: Thomson Reuters filed suit against ROSS back in 2020, alleging that it copied the entirety of the Westlaw database (after having been denied a license) to use as training data for its competing generative AI-powered legal research platform. In response to Reuters’ suit, ROSS argued, in part, that its unauthorized copying/use of the Westlaw database amounts to fair use. Specifically, ROSS claimed that it took only “unprotected ideas and facts about the text” in order to train its model; that its “purpose” in doing so was to “write entirely original and new code” for its generative AI-powered search tool; and that there is no market for the allegedly infringed Westlaw content consisting of headnotes and key numbers.
A Striking Win for Reuters
In what is being characterized as the first fair use-focused decision in the realm of copyright and AI, Judge Bibas of the U.S. District Court for the District of Delaware stated in his February 11 opinion that ROSS had directly copied 2,243 of Reuters-owned Westlaw’s headnotes, making it liable for copyright infringement. On this front, the court addressed two key elements of copyright infringement:
> Validity of Copyright: Judge Bibas affirmed his earlier determination that Westlaw’s headnotes are sufficiently original, as they are not merely copied judicial text but involve editorial judgment in selecting and phrasing legal principles, and thus, qualify for copyright protection. However, he noted that “if Thomson Reuters chooses to try this case based on a theory of infringement of individual headnotes as individual works rather than infringement of the compilation as a whole, there is still a factual dispute about which individual headnotes are both within the period covered by Thomson Reuters’s registrations and not in the public domain.”
> Copying and Substantial Similarity: By analyzing thousands of headnotes against ROSS’s AI training data, the court found undeniable similarity, concluding that ROSS had effectively used Thomson Reuters’ materials as a foundation for its competing legal research tool.
The Court’s Findings on Fair Use
Turning to ROSS Intelligence’s fair use defense, in furtherance of which it argued that its use of headnotes was transformative and necessary for AI innovation, the court was unpersuaded. Key takeaways from court’s application of the four-factor fair use test include …
> Purpose and Character: The court found that ROSS’s use was commercial and non-transformative. Unlike cases where fair use has been granted for technological innovation—such as Google’s copying of Java APIs—the court found ROSS’s copying was not necessary for interoperability and instead replicated Westlaw’s core functionality.
> Nature of the Copyrighted Work: While legal texts have factual elements, the court held that the editorial work involved in crafting headnotes involved sufficient creativity to warrant protection, though it acknowledged that they are not as creative as purely fictional works.
> Amount and Substantiality of Use: The court ruled in favor of ROSS on this factor, noting that while thousands of headnotes were used, the AI tool did not directly output headnotes to consumers.
> Market Harm: The court ruled in favor of Reuters, stating that ROSS’s use undermined existing and potential markets for AI-driven legal research. The ruling emphasized that this included not just Westlaw’s subscription service, but also potential licensing of its data for AI training.
Implications for Copyright & AI
This ruling could prove to be an influential one in the intersection of AI and copyright law, a nexus where litigation is currently quite rampant. Judge Bibas’ determinations seem to underscore that even in the era of AI, copyright protections remain robust, particularly when creative judgment is involved. The case could set a precedent that AI developers cannot freely use proprietary datasets without licensing agreements, though the full impact of the ruling will depend on whether it is appealed and if – and how – other courts interpret it.
The case is Thomson Reuters Enterprise Centre v. ROSS Intelligence Inc., 1:20-cv-00613 (D. Del.).