Anthropic must pay book authors $1.5 billion under a copyright settlement approved by a federal court in San Francisco after the company downloaded books from the piracy databases LibGen and PiLiMi between 2021 and 2022.

According to the court record, about 482,460 works were listed in the settlement, and 91.3 percent were claimed. The payout amounts to roughly $3,000 per claimed work, about four times the statutory minimum. Anthropic is also required to destroy the pirated files.

The authors retain claims related to AI outputs that reproduce original works, as well as claims involving Anthropic's future conduct. The settlement is the largest copyright settlement in class action history.

The payment addresses the piracy-related claims, not AI training itself. Judge Alsup had previously ruled that training AI systems on legally obtained books is "transformative - spectacularly so" and falls under fair use. The court document is available at Justia.

The case is one of several high-profile copyright disputes involving AI companies. The New York Times, individual authors, and rights holders have filed lawsuits against OpenAI, Meta, and others over the use of copyrighted works in model training. However, the Anthropic settlement is notable because it centers on the company's decision to obtain books from known piracy repositories rather than through purchase or licensing, a distinction that shaped both the court's analysis and the size of the settlement.

Whether large-scale scraping of internet content without authors' consent qualifies as legal acquisition remains unresolved, leaving the broader fair use dispute over AI training data open. The ruling is still a significant development for AI labs that trained models on web content without website owners' consent, a major source of training data for the industry.