In first published decision addressing use of copyrighted materials to train artificial intelligence, district court denies cross-motions for summary judgment, holding that it could not rule as matter of law that legal research startup’s alleged use of Westlaw headnotes to train a competing artificial intelligence platform constitutes fair use.
Thomson Reuters is the owner of Westlaw, the widely used legal research platform. Ross Intelligence is a legal research startup seeking to build a competing natural language research platform, designed with artificial intelligence (AI), that would allow users to ask legal questions and generate answers pulled from judicial opinions. To train its AI, Ross hired LegalEase Solutions (LegalEase), a third party, to create memos with legal questions and answers to be fed into Ross’ platform. LegalEase allegedly relied on Westlaw’s headnotes and Key Number System in doing so, in particular to draft the legal questions addressed in these memos. Thomson Reuters brought suit, asserting claims for copyright infringement and tortious interference with contract.
The parties cross-moved for summary judgment on various issues, including whether Ross infringed Thomson Reuters’ copyright through its use of Thomson Reuters’ materials in training its AI platform, whether any use by Ross of Thomson Reuters’ materials constituted fair use, whether Ross engaged in tortious interference with contract and whether Thomson Reuters’ tortious interference claim is preempted by the Copyright Act.
Turning first to the copyright infringement claim, the court held that there were triable issues of fact as to multiple aspects of Thomson Reuters’ claim. The validity and breadth of copyright protection were questions for the jury because there were factual disputes concerning the originality of both the Westlaw headnotes and Key Number System. Specifically, the Key Number System’s protectability depends on the extent to which the arrangement of the system involved original expression, while the headnotes’ protectability depends on the extent to which the headnotes mirror the content of uncopyrightable judicial opinions. Whether the content in memos used by Ross was substantially similar to the Westlaw materials likewise required a jury determination because it was disputed whether the similarities were due to the copying of Westlaw’s protected expression or to their common reliance on judicial opinions. By contrast, actual copying could be decided as a matter of law, because there was undisputed evidence that LegalEase had access to Westlaw and its headnotes and that it did, in fact, copy portions of the headnotes.
With respect to Ross’ fair use defense, the court held that there were triable issues of fact as to each of the four fair use factors. As for the first factor—the purpose and character of the use—the court relied heavily on case law concerning so-called “intermediate copying,” where “users copied material to discover unprotectable information or as a minor steps towards developing an entirely new product.” Thomson Reuters argued that the intermediate copying case law was inapt because “Ross simply sought to ‘train its AI’ by ‘cop[ying] the creative decisions of West[law]’s attorney-editors,’” with Ross arguing that “its AI studied the headnotes and opinion quotes only to analyze language patterns.” Given the factual dispute surrounding how Ross used the Westlaw content, the court ruled that it was a jury question whether “Ross’ AI only studied the language patterns in the headnotes to learn how to produce judicial opinion quotes” or “Ross used the untransformed text of headnotes to get its AI to replicate and reproduce the creative drafting done by Westlaw’s attorney-editors.”
Turning to the second factor—the nature of the copyrighted work—the court found that while this factor appeared to favor fair use, a jury trial was still needed. More specifically, the court found that Thomson Reuters’ Key Number System existed “far from the core of copyright”—which militated in favor of fair use—while the degree of protection afforded to the headnotes depended on “how closely [the] headnotes reflect the language of judicial opinions.”
The third factor—the amount and substantiality of the copying—also depended on factual disputes. While direct copying of Westlaw’s headnotes would counsel against fair use, if “Ross’ AI works the way that it says, it is likely fair use because it produces only the opinion, not the original expression.” Moreover, there were disputed factual questions as to whether “the scale of copying (if any) was practically necessary and furthered [Ross’] transformative goals.”
Finally, the fourth factor—the effect of the use upon the potential market for the copyrighted work—also could not be decided on summary judgment. While it was undisputed that the parties competed in the market for legal research platforms, the substitutability of their platforms depended on whether Ross’ use of Thomson Reuters’ material was transformative. If a jury found that the use was transformative, Ross’ platform would not be a market substitute. Moreover, the parties disagreed as to whether Ross’ use of Thomson Reuters’ training data harmed its ability to profit from that data. That too required a jury’s assessment. And because the fourth factor also takes into account “the public benefits the copying will likely produce,” a jury was needed to determine the benefit to the public of “allow[ing] AI to be trained with copyrighted material.” Accordingly, the parties’ cross-motions were denied as to Ross’s fair use defense.
As for the other two tortious interference claims—the first of which prohibited the use of any “spyware, malware” or similar software, and the second of which prohibited sharing passwords—the court held that there was no preemption because Section 106 has nothing to say about either of those issues. However, summary judgment was still improper on both claims because of factual disputes regarding whether (1) Ross had knowledge of the substance of Westlaw’s contract with LegalEase, (2) Ross knew that LegalEase was going to breach that contract and (3) Ross acted without justification.
As a final matter, the court granted summary judgment to Thomson Reuters on a variety of miscellaneous affirmative defenses. Laches was unavailable as a defense because it does not apply to copyright claims and Thomson Reuters did not delay in bringing its tortious interference claim, and the other defenses failed because they were largely duplicative of arguments raised in response to the copyright infringement and tortious interference claims.
Summary prepared by Wook Hwang and Edward Delman