In a landmark decision dated February 11, 2025, the United States District Court for the District of Delaware (Thomson Reuters Enterprise Centre GmbH and West Publishing Corp. v. Ross Intelligence Inc., No. 1:20-cv-613-SB) provides crucial guidance on the use of copyrighted works for artificial intelligence training.
The dispute pits Thomson Reuters, publisher of the Westlaw legal database, against Ross Intelligence. After being denied a license by Thomson Reuters, Ross obtained approximately 25,000 "Bulk Memos" through LegalEase to train its competing legal AI. These memos incorporated Westlaw's headnotes, original legal summaries synthesizing key points of law from judicial decisions. Judge Bibas, revising his 2023 decision, finds direct infringement of 2,243 headnotes.
The decision's major contribution lies in its thorough fair use analysis. The Court first rejects the transformation argument, stating that: "Ross's use is not transformative because it does not have a further purpose or different character from Thomson Reuters's". This means the Court considers Ross's use of headnotes to train its AI serves exactly the same purpose as Thomson Reuters: providing an efficient legal research tool.
The Court distinguishes this case from Google and Sony precedents, noting that: "intermediate copying cases are (1) computer-programming copying cases; and (2) depend in part on the need to copy to reach the underlying ideas. Neither is true here." In practical terms, the Court explains that Ross, unlike Google or Sony, did not need to copy the headnotes to access the underlying legal ideas, which were available in public court decisions.
Regarding market impact, deemed "undoubtedly the single most important element of fair use", the Court develops a two-pronged analysis. It first examines the effect on the primary market: "Even taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute", indicating that Ross deliberately sought to compete with Westlaw using a substitutable product. The Court then extends its analysis to the derivative market, stating that "the effect on a potential market for AI training data is enough", thus recognizing the existence of a specific market for AI training data worthy of protection.
Finally, the Court dismisses the public interest argument, noting that "the public's interest in the subject matter alone is not enough". This formulation means that while access to law serves a public interest, it does not justify copying Thomson Reuters' legal analysis, which represents protectable intellectual work.
This decision thus establishes a fundamental distinction between using protected works for training non-generative AI in a commercial context and the exceptions allowed for computer code. It also recognizes the emergence of a specific market for AI training data, now protected by copyright law. Given the significant implications for the AI industry, an appeal seems likely, especially as the Court specifies that its decision only concerns non-generative AI.
