On February 11, 2025, a Federal Judge issued summary judgement in one of the plethora of cases involving Artificial Intelligence, copyright infringement and fair use. The decision was surprising on several levels, not the least of which was the Judge’s apparent decision to sua sponte reconsider his earlier decisions in the case.
The case, Thomson Reuters Enterprise Centre GMBH v. Ross Intelligence, Inc., 1 involves the use of artificial intelligence to train a legal database, using copies of some 25,000 “headnotes” from the WestLaw legal research service. Ross first approached WestLaw for a license to their materials. Realizing that Ross intended to directly compete with the WestLaw service, they refused. As is typical of tech companies in these cases, Ross went ahead and used the material anyway, without license or payment. 2
Upon the release of the decision, most commentators focused on the fact that the fair use defense relied upon by the Defendant (and every other tech company in the world), was soundly rejected by the Court. But what precedential effect the case has moving forward is rather murky, mostly because of what was copied and the fact that the AI output was non-generative.
The headnotes at issue are nothing more than small snippets which summarize the rulings of the court in the case. The cases themselves, being government works, are not copyrightable. So, in order to properly summarize the cases, some degree of similarity is inevitable. Yet the Court broadly ruled that even where the headnote was a verbatim quote, from a non-copyrightable source, it was nonetheless copyrightable.
“More than that, each headnote is an individual, copyrightable work. That became clear to me once I analogized the lawyer’s editorial judgment to that of a sculptor. A block of raw marble, like a judicial opinion, is not copyrightable. Yet a sculptor creates a sculpture by choosing what to cut away and what to leave in place. That sculpture is copyrightable. 17 U.S.C. § 102(a)(5). So too, even a headnote taken verbatim from an opinion is a carefully chosen fraction of the whole. Identifying which words matter and chiseling away the surrounding mass expresses the editor’s idea about what the important point of law from the opinion is. That editorial expression has enough ‘creative spark’ to be original. (citation omitted). So all headnotes, even any that quote judicial opinions verbatim, have original value as individual works.” 3
Giving independent copyright status to a verbatim quote seems to me to be a terribly overbroad reading of the originality requirement. Especially where the Judge excluded from his ultimate decision all such verbatim quotes, rendering the above passage dicta. 4 But what it does explain is the ease in finding that 2,243 of the AI headnotes were substantially similar to the WestLaw headnotes.
This is important because without the finding of substantial similarity, the Court never gets to rule on the fair use defense, and why the case may be of limited value going forward. Because the AI at issue here creates output that is not generative. Asking the AI program for a summary of a Supreme Court case is not going to result in a generative output in the same way as asking for a painting of a blue unicorn on top of a mountain. There is but one Supreme Court case, and summarizing it is naturally going to result in some similarities. Unlike the blue unicorn, of which there are potentially thousands of possibilities and mountaintops, both real and fictional, which would surely outnumber those present here.
Nevertheless, kudos to the Judge to take the time to slog through 2,830 headnotes to conclude that 2,243 were substantially similar enough to be infringing. 5 But consider the impossibility moving forward where the AI program in question may have ingested millions, if not billions of pieces of artwork. How is a Court (or a jury) reasonably expected to make that comparison?
This is why the AI companies want to plant their flag on the shore of substantial similarity. The burden of showing substantial similarity from a few outputs generated from millions of potential sources is that daunting. Because remember, you don’t normally get to fair use if there is no substantial similarity. To counter this, the copyright owner needs to assert the reasoning of the 2nd Circuit in the TV Eyes case. 6
“By providing Fox’s content to TVEyes clients without payment to Fox, TVEyes is in effect depriving Fox of licensing revenues from TVEyes or from similar entities. And Fox itself might wish to exploit the market for such a service rather than license it to others. TVEyes has thus ‘usurp[ed] a market that properly belongs to the copyright-holder.’ (citation omitted) It is of no moment that TVEyes allegedly approached Fox for a license but was rebuffed: the failure to strike a deal satisfactory to both parties does not give TVEyes the right to copy Fox’s copyrighted material without payment.” 7
But back to the case at issue, what this case does show, is that the fair use defense is no longer the impregnable fortress that it once was.
For this, we begin to see the effect of the Warhol case. 8 As I predicted at the time, the holding did not bode well for AI developers. 9 Simply put, where the purpose of copying a work is the same as creating the original, this is not a “transformative” use that might tip the scales into fair use.
Here, it was plainly obvious that the AI database was built expressly to perform the same function as the WestLaw database.
“Ross was using Thomson Reuters’s headnotes as AI data to create a legal research tool to compete with Westlaw. It is undisputed that Ross’s AI is not generative AI (AI that writes new content itself). Rather, when a user enters a legal question, Ross spits back relevant judicial opinions that have already been written. (citation omitted) That process resembles how Westlaw uses headnotes and key numbers to return a list of cases with fitting headnotes.” 10
So, factor one goes to WestLaw. The Court awards factor two to Ross, as it does for factor three, in a muddled piece of reasoning that having read it three times still makes no sense.
But I digress.
But the negative impact on market effect is what carries the day for WestLaw.
“Even taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute. (citation omitted) And it does not matter whether Thomson Reuters has used the data to train its own legal search tools; the effect on a potential market for AI training data is enough. Ross bears the burden of proof. It has not put forward enough facts to show that these markets do not exist and would not be affected.” 11
So, moving forward, the problem for AI systems remains the same: Fair Use is not going to save you where the purpose of the works is the same and where they are intended to compete in the same marketplace with each other.
But potentially the most important factor may be that this case comes from the District Court in Delaware, where Getty Images has filed its class action suit against Stability AI. 12 If the WestLaw case is appealed (which is in doubt as the Defendant is no longer in business) 13 and affirmed, the lower court in the Stability AI case would be bound by it.
Dum Da Dum Dum.
Notes:
- 2025 WL 458520 U.S. District Court D. Delaware ↩
- Id. at 1 ↩
- Id. at 3 ↩
- Id. ↩
- Id. at 5 ↩
- Fox News Network, LLC v. TVEyes, Inc. 883 F.3d 169 (2d. Cir 2018) ↩
- Id. at 180 ↩
- Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 529–31, 143 S.Ct. 1258, 215 L.Ed.2d 473 (2023). ↩
- Supreme Court Saves the Derivative Works Right from “Transformative” Extinction; And Why AI Should Be Worried, “Lazy Appropriators” Beware: The “Warhol” Fair Use Standard Takes Hold ↩
- Ross at 7 ↩
- Id. at 10 ↩
- Getty Images lawsuit says Stability AI misused photos to train AI ↩
- Legal Research Company ROSS to Shut Down Under Pressure of Thomson Reuters’ Lawsuit ↩