Lucia Bautista
Delaware District Court Rules No Fair Use in Artificial Intelligence Case by Lucia Bautista
This past February, the United States District Court for the District of Delaware issued a landmark ruling in the case of Thomson Reuters v. Ross Intelligence. The dispute centered on Ross's unauthorized use of Thomson Reuters’s Westlaw headnotes. Ross was an AI-powered legal research platform that, after being denied a license to use Westlaw content directly, contracted with LegalEase to obtain training data in the form of “Bulk Memos” created using Westlaw headnotes.1
Judge Bibas reversed parts of his own 2023 ruling to find that Ross had directly infringed Thomson Reuters's copyrights in 2,243 headnotes. This finding of infringement was based on both actual copying and substantial similarity.2 The court determined that these headnotes reached sufficient originality to earn copyright protection even though they closely paraphrased judicial opinions—which themselves are not copyrightable.3 This case is one of many firsts providing critical insights into how courts may approach copyright issues in the age of artificial intelligence.
Most notably, the court rejected Ross’s fair use defense based on factors one and four. The court recognized that the headnotes were not highly creative works (fair use factor two) and acknowledged that the end product did not display the headnotes to the end user (fair use factor three).4 However, the court did find that the use was commercial and not transformative (fair use factor one) and that it harmed Thomson
1 Thomson Reuters Enter. Ctr. GMBH v. Ross Intel. Inc., No. 1:20-CV-613-SB, 2025 WL 458520, at 3 (D. Del. 2025).
2Id. at 12,14.
3Id. at 6–7; 17 U.S. Code § 102(b).
4 Thomson Reuters 2025 WL 458520, at 19–20.
Reuters’s existing and potential markets (factor four).5 This dispute, its analysis, and its outcome raise and respond to profound questions about the intersection of copyright law and AI development.
How should we distinguish between copyrightable expressions and uncopyrightable facts or ideas in the context of training data? This court held that even headnotes based on uncopyrightable judicial opinions contained sufficient creative expression to merit copyright protection.6
How does the present case compare with other notable software cases? Judge Bibas differentiates this issue from cases like Oracle and Sony, likens the present matter to Warhol and reasons that the “copying is [not] reasonably necessary to achieve the user’s new purpose.” 7
Should copyright law treat non-generative AI differently from generative AI? This court believes so, explicitly noting that this case only addressed non-generative AI–systems that retrieve existing content rather than create new text.8
What's the proper balance between creator and consumer interests when fostering innovation in AI? Judge Bibas emphasizes that copyright protection should incentivize the creation of valuable tools and benefit public interest by being accessible. However, he explains that because legal opinions are freely available, Ross’s legal research platform does not increase accessibility.9
As AI development continues to rapidly evolve, questions about the legal protection afforded and denied to these works grows more pressing. The Thomson
5Id. at 16 and at 21–23.
6Id. at 7.
7Id. at 19.
8Id.
9Id. at 22.
Reuters case furthers arguments that courts are unwilling to expand copyright law to accommodate AI training needs. In light of this opinion, companies developing AI systems should proceed with caution when using copyrighted materials, even if those materials rely heavily on facts and do not appear in the final product.