Facts
Thirteen authors, mostly well-known fiction writers, sued Meta for downloading their books from online "shadow libraries" and using them to train its Llama large language models. According to the order, Meta downloaded the Library Genesis (LibGen) database in October 2022 and Anna's Archive in early 2024, both by torrenting, and the parties disputed how far Meta re-uploaded what it downloaded. Both sides moved for partial summary judgment on fair use. The plaintiffs argued that Llama could reproduce snippets of their books and that Meta's unlicensed use had deprived them of the market for licensing their works as AI training data.
What was the legal question in Kadrey v. Meta Platforms, Inc.?
The question was whether copying copyrighted books without permission to train large language models is fair use under the four factors of section 107 of the Copyright Act. This raised how transformative the use was, what weight to give the fact that the books came from shadow libraries, whether a market for licensing works as AI training data is one the law protects, and whether the model's outputs could harm the market for the books through market dilution.
What did the US District Court for the Northern District of California decide?
Judge Chhabria denied the plaintiffs' motion and granted Meta's. Copying the books to train Llama was highly transformative, and neither its commercial purpose nor the use of shadow libraries tipped the first factor toward the plaintiffs. Both of the plaintiffs' market theories failed: even with adversarial prompting, Llama would not produce more than 50 words of any of their books, and they were not entitled to a market for licensing their works as AI training data. The potentially winning argument was market dilution by a flood of similar works, but the plaintiffs barely raised it and offered no evidence. On that record the fourth factor favoured Meta, so copying the books for training was fair use.
Why does Kadrey v. Meta Platforms, Inc. matter?
The ruling reached the same result as Bartz v. Anthropic, decided two days earlier in the same district, but on different reasoning. Judge Chhabria treated market dilution as potentially decisive under the fourth factor, predicted that unlicensed training will often be unlawful, and stressed that the ruling binds only these thirteen authors. It therefore does not establish that training is lawful in general and indicates what later plaintiffs would need to prove about dilution.
Related stages
The order states that summary judgment for Meta on the DMCA claim would follow in a separate ruling. The claim that Meta distributed the works while torrenting was not decided and the case continues. On 8 July 2026 the court denied the plaintiffs leave to seek an interlocutory appeal under 28 U.S.C. § 1292(b), noting that a final judgment would let the issue reach the court of appeals with others. A docket entry of 26 March 2026 set a dispositive motion hearing for 25 February 2027; some expert and briefing deadlines were extended in September 2026. See bartz-v-anthropic and thomson-reuters-v-ross.