On February 11, 2025, Judge Bibas of the Third Circuit, sitting by designation in the US District Court for the District of Delaware, issued a decision granting partial summary judgment to Thomson Reuters in the closely watched case of Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. Judge Bibas revisited his 2023 decision largely denying summary judgment in the case due to disputed facts,[1] and the new decision is the first time a U.S. court has reached a conclusion concerning the application of copyright’s fair use doctrine to the use of works in training data for artificial intelligence (AI) and machine learning. While the case does not involve generative AI, but rather a more traditional application of AI, and fair use analyses are always fact-specific, the court’s decision rejecting a fair use defense is an important development.
The case was brought by Thomson Reuters, the owner of the legal research platform Westlaw, against Ross Intelligence, a start-up company that sought to provide a competitive legal research platform powered by artificial intelligence.[2] Ross’s idea was to create a “natural language search engine” that would allow users to enter legal questions and receive as answers quotations from judicial opinions selected by AI.[3]
Judicial opinions in Westlaw are accompanied by “headnotes” that “summarize key points of law and case holdings” and are organized by “Key Number.”[4] Initially, Ross sought to license Westlaw’s content to train its AI search function, but Thomson Reuters declined.[5] As a substitute for training on Westlaw content, Ross commissioned approximately 25,000 so-called “Bulk Memos” from a third-party legal research company. The Bulk Memos were “compilations of legal questions with good and bad answers.”[6] However, the Bulk Memos were created using Westlaw headnotes and in many cases contained language similar to Westlaw Headnotes.[7] Thomson Reuters sued Ross for copyright infringement because of its use of Westlaw content in training its AI technology.[8] In 2023, the court largely denied motions for summary judgment, but in the run-up to trial, the court invited the parties to renew their summary judgment briefing.[9]
Based on the new briefing, the court granted Thomson Reuters summary judgment of direct infringement with respect to 2,243 headnotes, while leaving for trial disputes as to thousands more headnotes, additional editorial content, and Thomson Reuters’s Key Number System.[10] In doing so, the court rejected all of Ross’s asserted defenses, including most notably fair use.”[11]
I. Direct Copyright Infringement and Miscellaneous Defenses
Before getting to fair use, the court first considered whether Thomson Reuters had a prima facie copyright claim against Ross and whether other defenses were available to Ross.
First, the court found that Thomson Reuters owned valid copyrights, although it allowed disputes as to the originality of specific headnotes to go to trial.[12] The court explained that the originality threshold is “extremely low” and requires only “some minimal degree of creativity,” a requirement that can be met by distilling a judicial opinion into a headnote.[13] Additionally, the court found that the Westlaw Key Number System, a “taxonomy” that Thomson Reuters uses to “organize its materials,” is sufficiently original to be protected by copyright.[14]
Second, the court analyzed whether Thompson Reuters had shown actual copying of Westlaw headnotes and substantial similarity between the headnotes and the Bulk Memos used as training data. In addition to relying on a concession by Ross’s expert, the court independently compared 2,830 Bulk Memo questions with the corresponding headnotes and judicial opinions and found strong circumstantial evidence of copying of 2,243 headnotes.[15] The court also found substantial similarity between the original expression in those headnotes and the Bulk Memos, because the language of those headnotes “very closely tracks the language of the Bulk Memo question but not the language of the case opinion.”[16]
Finally, the court gave short shrift to Ross’s defenses of innocent infringement,[17] copyright misuse,[18] merger,[19] and scenes à faire.[20]
II. Fair Use
The decision is particularly notable for finding that Ross’s copying of Westlaw headnotes in AI training data was not a fair use. This is a reversal from the court’s 2023 ruling denying summary judgment on fair use because of disputed facts, including particularly whether Ross’s AI training was transformative.[21] Reexamining the issue, and noting that fair use is a mixed question of law and fact, the court found that the disputes between the parties did not involve questions about historical facts, intentions and predictions, but rather how to apply the law to such facts, making the issue ripe for summary judgment.[22] The court thus proceeded to consider each of the four statutory fair use factors in turn, finding that the important first and fourth factors favored Thomson Reuters, while the less important second and third factors favored Ross. Weighing the factors together, the court granted summary judgment for Thomson Reuters on fair use.
Purpose and character of the use
The first factor, “the purpose and character of the use,”[23] has often had a determinative effect in fair use analysis, and the court found that this factor weighed in favor of Thomson Reuters.[24]
Ross admitted that its copying was commercial. However, as explained in the US Supreme Court’s 2023 Warhol decision,[25] that fact needs to be balanced against the degree of difference between the purpose and character of the original work and that of the secondary use.[26] The court determined that “Ross’s use was not transformative because it did not have a ‘further purpose or different character’ from Thomson Reuters’s.”[27] Rather, “Ross was using Thomson Reuters’s headnotes as AI data to create a legal research tool to compete with Westlaw.”[28] Specifically, because “Ross spits back relevant judicial opinions that have already been written,” similar to how Westlaw uses headnotes and the Key Number System to help researchers find judicial opinions, the functions are similar.[29]
Ross had argued that Westlaw headnotes are not displayed as part of its final product, and thus that its use of the headnotes was a permissible “intermediate step” in the creation of its product.[30] In doing so, Ross relied on a line of cases finding that intermediate copying of computer programs is permissible when necessary to achieve technological compatibility between an innovative product and an existing one. The court distinguished those cases, finding that their treatment of computer programs does not necessarily apply to the written words of legal text, and that here, copying was not “reasonably necessary to achieve the user’s new purpose.”[31]
Because the purpose and character of Ross’s copying was to “make it easier to develop a competing legal research tool,” the court concluded that Ross’s use was not transformative.[32] However, the court went out of its way to note that “only non-generative AI is before me today.”[33]
Nature of the copyrighted work
By contrast, the court found that the second factor, “the nature of the copyrighted work,”[34] favored Ross. The court observed that while creation of the headnotes required enough creativity to give rise to a valid copyright, “that creativity is less than that of a novelist or artist drafting a work from scratch.”[35]
Amount and substantiality of the portion used
The court also found that the third factor, “the amount and substantiality of the portion used in relation to the copyrighted work as a whole,”[36] favored Ross.[37] Importantly, the court explained that “[w]hat matters is not ‘the amount and substantiality of the portion used in making a copy, but rather the amount and substantiality of what is thereby made accessible to a public for which it may serve as a competing substitute.’”[38] The court reasoned that because “Ross did not make West headnotes available to the public, Ross benefits from factor three.”[39]
Effect of the use on the potential market for or value of the copyrighted work
Finally, the court examined the fourth fair use factor, “the effect of the use upon the potential market for or value of the copyrighted work,”[40] finding that it favored Thomson Reuters.[41] In doing so, the court focused on the fact that Ross “meant to compete with Westlaw by developing a market substitute,” and also accepted the concern that Ross could affect a potential derivative market for licensing AI training data.[42] Addressing a “public interest” argument from Ross, the court determined that although there is “a public interest in accessing the law,” there is no public “right to Thomson Reuters’s parsing of the law.[43]
Balancing the four factors, the court found that the ones favoring Thomson Reuters mattered more than those favoring Ross, and so granted summary judgment for Thomson Reuters on fair use.[44]
III. Conclusion
As the court itself noted, this case involved use of AI to provide a search capability, not to generate new material. Perhaps the most interesting takeaways here are the court’s willingness to consider the totality of Ross’s product plans, and its assessment that those plans involved a purpose similar to and competitive with Westlaw headnotes, rather than focusing narrowly on the immediate purpose of Ross’s copying. This approach could be influential to courts considering fair use issues involving generative AI systems.
Footnotes
[1] Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., 694 F. Supp. 3d 467 (D. Del. 2023).
[2] Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., No. 20-0613, Mem. Op. at 2-3 (D. Del. Feb. 11, 2024) [hereinafter “Mem. Op.”].
[3] 694 F. Supp. 3d at 475.
[4] Mem. Op. at 2.
[5] Id. at 3.
[6] Id.
[7] Id.
[8] Id.
[9] Id.
[10] Id. at 9-10, 12-13, 23.
[11] Id. at 23. The court left intact its prior rulings finding disputed facts as to Ross’s potential contributory and vicarious liability for the actions of the contractor that created the Bulk Memos, and Ross’s potential tortious interference with the contract providing that contractor access to Westlaw.
[12] Id. at 5-6.
[13] Id. at 6-8 (noting that the “degree of overlap between the headnote text and the case opinion” is not “dispositive of originality.”).
[14] Id. at 8.
[15] Id. at 12 (finding that the Bulk Memo questions were more similar to the headnotes than the underlying opinions).
[16] Id. at 14.
[17] Id. (innocent infringement can only limit damages, and does not apply if a work bears a copyright notice, as Westlaw does).
[18] Id. at 15 (Ross did not show that “Thomson Reuters misused its copyrights to stifle competition”).
[19] Id. (“there are many ways to express points of law from judicial opinions”).
[20] Id. (“nothing about a judicial opinion requires it to be slimmed down to Thomson Reuters’s headnotes or categorized by key numbers”).
[21] 694 F. Supp. 3d at 482-87.
[22] Mem. Op. at 16.
[23] 17 U.S.C.§ 107(1).
[24] Mem. Op. at 16.
[25] Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 525 (2023).
[26] Mem. Op. at 16-17.
[27] Id. at 17 (quoting Warhol, 598 U.S. at 529).
[28] Id.
[29] Id. at 17.
[30] Id. at 17-18.
[31] Id. at 18-19 (quoting Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 532 (2023)).
[32] Id. at 19.
[33] Id.
[34] 17 U.S.C.§ 107(2).
[35] Mem. Op. at 20.
[36] 17 U.S.C.§ 107(3).
[37] Mem. Op. at 20-21.
[38] Id. at 21 (quoting Authors Guild v. Google, Inc., 804 F.3d 202, 222 (2d Cir. 2015); emphasis in original).
[39] Id.
[40] 17 U.S.C.§ 107(4).
[41] Mem. Op. at 21-23.
[42] Id. at 21-22.
[43] Id. at 22-23.
[44] Id. at 23.
[View source.]
