Facts
Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued Anthropic PBC, the developer of the Claude models, for copyright infringement in the U.S. District Court for the Northern District of California. According to the order, Anthropic downloaded more than seven million pirated copies of books from pirate sites, and also bought print books, stripped their bindings and scanned every page. It kept these copies in a central digital library it planned to retain indefinitely and selected sets of books from that library to train its large language models. Anthropic moved for summary judgment on fair use.
Question
The court had to decide whether each distinct use of the books was a fair use under Section 107 of the Copyright Act. It separated three uses: copying books to train large language models, converting lawfully purchased print books into digital copies for the library, and downloading and retaining pirated copies to build a permanent, general-purpose library. A central question was whether the transformative character of the later training use could also excuse the way the copies had been acquired and kept.
Decision
Judge William Alsup granted summary judgment to Anthropic on the training use, describing it as exceedingly transformative and a fair use. The print-to-digital conversion of purchased books was also fair use, because Anthropic merely replaced print copies it owned with more convenient digital ones without adding new copies or redistributing them. The court denied summary judgment on the pirated library copies. Building a permanent library from pirated books was not itself a fair use, and a later purchase of the same book would not absolve Anthropic of liability, though it could affect statutory damages. The court ordered a trial on the pirated copies and damages.
Why it matters
The order was among the first to hold that training a generative model on lawfully acquired books can be fair use. It assessed how training data is obtained and stored separately from the training itself. That distinction exposed Anthropic to potential statutory damages for millions of pirated works and shaped the settlement that followed. Two days later, another judge in the same district placed different weight on market harm in Kadrey v. Meta.
Related stages
The parties later agreed a $1.5 billion class settlement. Judge Alsup granted preliminary approval on September 25, 2025, and Judge Araceli Martínez-Olguín granted final approval on July 20, 2026 (Dkt. 680). The settlement covers 482,460 works, at about $3,000 per work before fees and costs. In Kadrey v. Meta (No. 23-cv-03417-VC, June 25, 2025), Judge Chhabria granted Meta partial summary judgment on fair use because the plaintiffs had not shown market harm, while stressing that the ruling did not mean Meta’s use was lawful.