Anthropic agrees $1.5 billion settlement over pirated books

Anthropic, the U.S. AI firm behind the Claude model, reached a $1.5 billion settlement with authors and publishers accused of using pirated e‑books from Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) to train its system. The Northern District of California approved the deal on July 20, covering roughly 480,000 works at about $3,000 per title, with attorney fees of about $115 million. The court distinguished the illegal acquisition of digital copies from Anthropic’s practice of buying physical books, removing their bindings, scanning each page and destroying the originals, which it deemed fair use because the scans replaced the physical format and were not distributed.

The settlement does not address Anthropic’s internal “Project Panama,” a program that bulk‑purchased and digitized millions of printed books for internal use. While the court ruled the scanning of lawfully bought books was permissible, it affirmed that downloading and retaining pirated e‑books remains unlawful, setting a precedent for how AI companies must source training data.