Copyright Fair Use for GenAI Training Data – Summary Judgment Decision Arriving Soon?

One of the most impactful copyright law issues of our time is whether the unauthorized use of copyrighted materials as training data for generative artificial intelligence (GenAI) models is protected fair use under the Copyright Act. A substantive ruling on this issue could arrive soon.

While the most talked about GenAI litigations are still at the pleading stage, one case—squarely addressing fair use for GenAI training data—is already awaiting a summary judgment ruling.

In Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc., 20-cv-00613 (D. Del.), the parties filed cross-motions for summary judgment on defendant’s affirmative defense of fair use. Those motions have been fully briefed since February 2023—nearly four months ago. 

The briefing is partially redacted, but the essence of the arguments is mostly clear. The case involves alleged unauthorized use of proprietary content from Thomson’s Westlaw legal research database as training data for ROSS’s competing GenAI legal research tool. The facts are unique compared to some other pending GenAI cases in that ROSS is alleged to have induced a third-party Westlaw licensee to obtain the data for ROSS after ROSS had been denied its own license. As such, the facts differ from other GenAI cases dealing with scraping publicly available data from the web. In any event, ROSS asserts that its use of Westlaw content (however it was acquired) as training data for its GenAI model is protected fair use. 

Among other fair use arguments, ROSS argued under the first fair use factor that it took only “unprotected ideas and facts about the text” in order to create its GenAI model, which is protected “research” under section 107 of the Copyright Act. Further, ROSS argued, its “purpose” was to “write entirely original and new code” for its GenAI search tool. ROSS also argued, under the fourth factor, that there is no market for the allegedly infringed Westlaw content consisting of headnotes and key numbers. (Dkt. 287, 339, 377.)

On the other hand, Thomson argued that ROSS’s “purpose” under the first fair-use factor was to supplant Westlaw with a competing, commercial legal research tool. Under the fourth factor, Thomson argued, inter alia, that ROSS harmed the market for Westlaw subscriptions (in which licensees pay to access the content), as well as the potential market for licensing of the content to third parties as training data for other GenAI tools. (Dkt. 297, 330, 370.)

The parties also recently filed “supplemental authority” briefs regarding the U.S. Supreme Court’s decision in Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. ___, 143 S. Ct. 1258 (2023) regarding transformative use under the first fair-use factor. 

According to Thomson, the Goldsmith decision confirms that: (1) “when the parties make . . . similar uses of content, it weighs against fair use”; (2) “when combined with the fact that a use is not transformative, commercialism weighs against a finding of fair use unless the use is otherwise justified”; and (3) “where there is no justification, the fact that the use was both commercial and for a highly similar purpose as the original weighs against fair use.” (Dkt. 494.) 

ROSS responded that Goldsmith “has limited applicability here” because (1) unlike here, “the use of the original and the allegedly infringing images in Goldsmith were identical; they were both used as illustrations in magazine articles”; (2) Goldsmith did not alter the Supreme Court’s approach to commercialism under fair use factor one, but rather reaffirmed its earlier decisions holding “that when uses are highly transformative, commercialism is less important”; and (3) search functionality is not protected by copyright but rather “is the very innovation that fair use protects.” (Dkt. 495.) 

As commentators continue to puzzle over what Goldsmith means for GenAI, at least in this particular case, on these facts, Thomson (arguing against transformative fair use) seemed to put more stock in Goldsmith than did ROSS (arguing in favor of fair use).

Discovery and briefing continue on other issues in Thomson Reuters, and it is unclear when the court will rule on the parties’ pending summary judgment motions on fair use. 

Of course, it is possible that Thomson Reuters will not be the first case to decide this issue, or that a summary judgment ruling could be tightly limited to its facts, or simply find genuine issues of fact for trial without much substance. But, for now, the case appears to be the lead horse by some distance and carries the possibility of a far-reaching dispositive ruling, meriting a close watch.