Meta wins narrow fair use battle on behalf of AI developers, copyright holders get a roadmap for future cases

What happened?

On June 25 and June 27, 2025, Judge Chhabria issued two orders in Richard Kadrey, et al. v. Meta Platforms, Inc., No. 3:23-cv-03417-VC (N.D. Cal. June 25, 2025), dismissing certain allegations by 13 best-selling authors that the use of copyright protected works to train Llama, Meta’s Artificial Intelligence Large Language Model (LLM), constituted copyright infringement. The first order dismissed the copying claims, while the second order dismissed the DMCA claims. The case will continue with copyright claims related to Meta’s alleged downloading of pirated, “shadow libraries.” The focus of this post is the first order. 

Key takeaways:

• The Court noted throughout the order that copyright holders could show market harm to defeat a fair use defense and although these plaintiffs failed to develop a record that the Court could rely on, the order is somewhat instructive to future plaintiffs.
• While the Court granted Meta’s partial motion to dismiss, the order was limited to the plaintiff’s specific market harm argument which the Court described as “so weak that it does not move the needle[.]”

Summary of facts:

Authors of certain books sued Meta for using their copyrighted books to train Llama. Despite initial attempts at licensing these books to train Llama, Meta ultimately trained Llama using copies of books downloaded from “shadow libraries.”

Analysis:

Looking at the four statutory fair use factors, the Court noted that while training LLMs is undoubtedly transformative and favors defendants, it is only one factor in the fair use analysis. Importantly, harm of the market for the copyrighted work is “undoubtedly the single most important element of fair use,” and “the only harm relevant is the harm of market substitution.”   The Court, being very critical of plaintiffs’ arguments and the record in the case stated:

“In cases involving uses like Meta’s, it seems like the plaintiffs will often win, at least where those cases have better-developed records on the market effects of the defendant’s use. No matter how transformative LLM training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books. And some cases might present even stronger arguments against fair use.”

Nonetheless, the Court analyzed three types of potential market harm:

(1) The AI model will regurgitate protected works: Creating free substitutes to the copyright protected works that harm the market.

Although the plaintiffs alleged generally that Llama could harm the market for protected works by generating exact or near-exact replicas of their works, the Court was persuaded by Meta guardrails which did not allow Llama to reproduce any meaningful portion of the plaintiffs’ books.  The Court noted that Llama was not able to regurgitate more than 50 words from any of the plaintiffs’ books, and so it would have no significant effect upon the potential market for the books.

(2) The market for licensing works for training AI.

Although the authors alleged that Meta’s actions harmed the potential market that could exist for licensing works specifically for training LLMs, the Court concluded that even if such market exists, which it does not, the market is not one that the plaintiffs are legally entitled to monopolize if the fair use defense applies. The Court explained, in every fair use case, the “plaintiff suffers a loss of a potential market if that potential market is defined as the theoretical market for licensing the use.”

(3) The market harm due to similar substitutes.

While the authors did not address this potential harm, the Court explicitly stated a strong record and argument on these grounds likely would have defeated any fair use argument. The Court suggested that people can or will soon be able to use LLMs to generate massive amounts of content in significantly less time and with less effort than it would take an author. This in turn would create a substitute market for creative works which would compete directly with those of the authors.  The Court noted that Amazon already appears to be flooded with AI-generated books, and that this harm could win the fair use issue:

“Harm from this form of competition is the harm of market dilution […] or indirect substitution. […] Indeed, it seems likely that market dilution will often cause plaintiffs to decisively win the fourth factor—and thus win the fair use question overall—in cases like this.”

The practical takeaway:

Judge Chhabria lays out the strongest potential arguments for copyright holders on the fair use issue about using copyright-protected works to train LLMs. While a lot of attention is paid to the whether the use is transformative, as the Court noted, market harm is the most important of the fair use elements and an LLMs ability to generate content that directly completes with the copyright holders will likely be an important, and likely dispositive, issue for future cases.