On May 1, 2025, Judge Vincent Chhabria held a hearing on the respective summary judgment motions of the parties in Kadrey v. Meta, a copyright infringement suit brought by book authors against Meta. If there’s one AI lawsuit to watch, this is it. Forget about the New York Times’ case, at least for now.
The parties have vastly different views of what alleged facts are even amenable to summary judgment. Banking on the notoriety of “pirated” books sites and illegal file sharing, Kadrey argues that only Meta’s downloading and file sharing of “pirated” books can be decided as infringement on summary judgment (and Meta’s training of its model should go to trial because of disputed facts). By contrast, Meta argues that everything about the copyright claim (except for the alleged unauthorized distribution of works by Meta through file sharing) can be considered and decided on summary judgment as a part of its fair use defense, including the downloading of the works. We created an infographic to understand the 2 different partial summary judgment motions, with links to their respective briefs and replies.
After 2 hours and 38 minutes of a spirited oral argument, Judge Chhabria adjourned the hearing. The Judge grilled both sides with tough questions. Meta’s lawyer, the Supreme Court lawyer Kannon Shanmugam, went first. Then Kadrey’s lawyer, David Boies.
It’s foolish to predict the outcome from an oral argument. That hasn’t stopped me before.
With that caveat, I think Meta’s attorneys have to be pleased. Here’s why.
1. Judge Chhabria appeared inclined to conclude that Meta’s use of copyrighted works to train an AI model is a transformative use. He asked far fewer questions on Factor 1 of fair use (the purpose and character of the defendant’s use of the plaintiffs’ works)–and point blank said to Shanmugam that he was inclined to agree that such use was a transformative, even a “highly transformative” use. (His 12 questions before the hearing raised 2 different bases for characterizing the purpose of Meta’s use as transformative: developing a new technology including what it is capable of producing or doing something “different from a human reading the book” focusing “simply” on “training the language model with it, without regard to what the language model will ultimately create.” It’s unclear if the Judge favored one formulation, but since most of his questions at oral argument were about the flooding the book market theory of harm, that suggests the inclusion of what the AI is capable of producing is relevant to the purpose of the defendant’s use.)
2. Judge Chhabria appeared inclined to reject the Kadrey book authors’ argument that lost licensing in itself would establish market harm under Factor 4. In questioning both sides, Judge Chhabria was quite skeptical that Kadrey’s asserted lost licensing in itself would establish market harm under Factor 4, given the inherent circularity problem that, in every case, the plaintiff will assert lost licensing (indeed, that’s one of the common forms of actual damages). Boies attempted to convince Judge Chhabria that licensing the works is an important right of the copyright holder.
3. Most of the 2 and a half hours of oral argument were spent on one concern: from the start of the questioning of Meta’s attorney, Shanmugam, Judge Chhabria focused on Factor 4 and the potential theory of market harm that he considered, if proven, to weigh strongly against fair use. The Judge called Factor 4 the “most important factor,” citing the dicta from Harper & Row. Both from his list of 12 Questions and from his questions at the hearing, Judge Chhabria appears inclined to view the issues in the AI cases (“zooming out” from the particular record in this case) as presenting a highly transformative use to develop AI but also a potential market harm at least under one theory. That was the reason why he asked counsel if they could name a case in which “the secondary use was found to be transformative but the fair use doctrine was found not to apply (perhaps because of the fourth factor).” (The answer is yes: Warner Brothers v. RDR Books, with the Harry Potter Lexicon ultimately being published after the publisher fixed its excessive copying.) Judge Chhabria seems inclined to recognize that Meta had a transformative purpose in using the copyrighted works to develop its AI model, but also inclined to recognize, if this argument was properly raised for summary judgment, a potential market harm from Meta’s production of a technology that could flood the market with competing works, at least generally speaking.
Judge Chhabria referred to this harm as an “obliteration” theory of market harm:
What if the copying of the copyrighted works to train AI models “lead[s] to the flooding of the market in copyrighted works that substantially diminishes the value in the works” used to train the models?…
How could that be fair use?
Companies can create a product capable of producing infinite number of competing works … and dramatically change and obliterate the market for the [original] works.
I don’t understand how that can be fair use.
Judge Chhabria
Yes, this latter thread sounds very, very bad for Meta. But it might not be.
During the questioning of Kadrey’s lawyer, Boies, Judge Chhabria devoted most of his questions–for over an hour—to whether Kadrey had even raised this theory of market harm or presented sufficient evidence to support it to survive summary judgment. The Judge also ruminated over the burden of proof on fair use, which is an affirmative defense. At one point, there was an exchange between the Judge and Boies over whether it was even permissible to introduce new evidence at trial that wasn’t presented in the summary judgment record and not disclosed during discovery. And the Judge also appeared to misinterpret Kadrey’s motion for summary judgment as covering Meta’s fair use defense for AI training. Boies clarified that it does not.
As to the question whether Kadrey had raised the issue of market obliteration, Boies argued that (1) Meta has the burden of proving an affirmative defense, and (2) Kadrey, in fact, raised this theory primarily through their expert, an economist, Daniel F. Spulber. See below. (Kadrey’s reply brief does briefly mention this theory of flooding the book market at p. 34, but Boies did not cite this brief during the hearing.) If my notes are correct, Boies even quoted the heading used in Spulber’s report at p. 77: “C. Meta Risks Depriving Plaintiffs of Revenues by Using Copyrighted Books as Inputs to Allow the Creation of Works that Compete with Plaintiffs’ Works.”
Boies also mentioned that some of the plaintiffs, including Christopher Farnsworth, had testified that Meta was able to generate outputs in their style and mimic their writings. At times, though, Judge Chhabria appeared to be asking for something more concrete to establish that Meta’s Llama will soon lead to this flooding of the book market to obliterate the plaintiffs’ books, although he also said to Meta’s lawyer that isn’t it common sense it will do so (at least in a general way).
Given this exchange on market harm, Judge Chhabria could rule that a trial is needed and reject both parties’ motions for summary judgment. And I could be way off in interpreting all the questions asked during nearly 3 hours of oral argument.
But, if Factor 4 and market harm from potential flooding the book market (not to mention, the potential public benefit from Meta’s technology) is really the main dispute to decide for fair use, that puts Meta one step away from prevailing on fair use. Even if the case goes to trial, Judge Chhabria will have the expected JMOL motions to decide after trial. And if he agrees that the use was highly transformative to develop Meta’s new AI model and that lost licensing is not in itself enough to establish market harm, that greatly reduces the fight to a battle (of experts) over the flooding the book market theory of harm, or what the Judge repeatedly referred to as “obliterating” the market for the original works because the technology can produce billions of comparable works en masse.
If you asked Meta at the start of the case, would they take their chances on disproving market harm and showing public benefit of Llama under Factor 4, I’m sure they would.
As for the plaintiffs’ DMCA CMI 1202(b)(1) claim, the Judge asked no questions at all. As for the plaintiffs’ claim of infringement based on leeching or sharing of files with third parties, there was surprisingly very little discussion at all. The Judge did ask some questions about what recognizing fair use from copies downloaded from “pirated” books sources or shadow libraries. Boies made several good arguments about how it would legitimize those sources with competitors and make them even more popular, whereas Shanmugam said it’s essentially just unauthorized copy of an unauthorized copy that shouldn’t affect the fair use analysis if the purpose is transformative. The Judge appeared to be concerned about the shadow libraries, but, if the number of questions is any indicator, this issue was secondary to the market harm issue discussed above.
DOWNLOAD KADREY’S ECONOMIST DANIEL SPULBER EXCERPTS OF REPORT:
Related Stories