Kadrey v. Meta Platforms, Inc.

Kadrey v. Meta Platforms, decided June 25th, 2025. Judge Vince Chhabria, in the Northern District of California, denied the plaintiffs' motion for partial summary judgment and granted Meta's cross-motion — holding that, on this record, Meta's copying of thirteen authors' books to train its Llama large language models was fair use. The court took pains to stress that the ruling is narrow. Here's the brief.

The plaintiffs are thirteen authors — mostly well-known fiction writers — who sued Meta for downloading their books from online "shadow libraries" and using those books to train Meta's generative AI models, its large language models, called Llama.

Meta obtained the books from repositories such as LibGen and Anna's Archive, in part by torrenting them, and used them as training data without permission from, or payment to, the copyright holders.

The authors brought several claims, including direct copyright infringement based on Meta's reproduction of their works, a claim under the Digital Millennium Copyright Act, and a separate claim that Meta unlawfully distributed their works during the torrenting process.

The parties filed cross-motions for partial summary judgment on fair use. The plaintiffs argued Meta's conduct could not possibly be fair use; Meta argued its copying was fair use as a matter of law. Meta did not dispute that the plaintiffs had established a facial case of infringement of their reproduction rights. This order resolves those cross-motions.

The question was whether Meta's reproduction of the plaintiffs' books to train Llama was protected by fair use — the affirmative defense codified in Section 107 of the Copyright Act.

Section one oh seven directs courts to weigh four factors: the purpose and character of the use, including whether it is commercial; the nature of the copyrighted work; the amount and substantiality of the portion used; and the effect of the use on the potential market for the work. Fair use, the court noted, is a flexible concept — the factors are not applied mechanically, and one may matter more in some contexts than others. Here, the court explained, the fourth factor, market effect, was the most important.

On the first factor, the court held Meta's use was highly transformative.

The purpose of the copying, the court reasoned, was to train large language models — innovative tools that generate diverse text and perform a wide range of functions distinct from reading a book for entertainment or education. Because that ultimate use was transformative, so too was the downloading that fed it.

Turning to market effect, the court addressed the plaintiffs' two theories: that Llama could reproduce small snippets of their books, and that Meta's use deprived them of a market to license their works as AI training data. The court rejected both — Llama could not generate enough text from the books to matter, and the authors were not entitled to a licensing market for training data.

The court then identified a third, potentially stronger theory it called market dilution: that a model trained on the books could flood the market with competing works and reduce demand for the originals. But the plaintiffs, the court found, barely developed it and offered no evidence.

With no genuine dispute of fact on dilution, the fourth factor favored Meta, and Meta was entitled to summary judgment on its fair use defense to the training claim. The court was emphatic about the limits of its ruling.

The decision holds only that, on this record, these thirteen plaintiffs failed to show that Meta's use of their books to train Llama was infringing. Because it is not a class action, the court noted, it affects only these authors' rights, and it does not hold that training on copyrighted works is lawful in general. The court suggested that in another case, with a developed record on market dilution, the same conduct might not be fair use. Two matters were left for separate resolution: the plaintiffs' DMCA claim, which the court said it would grant to Meta in a separate order, and their distribution claim, which remained live.

Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, decided June 25th, 2025. I'm John Goodhue. Thanks for watching.