Court rejects Anthropic's $1.5 billion settlement with authors in AI pirated books case

Federal Judge William Alsup has refused to approve a settlement between Anthropic and authors who accuse the company of using pirated copies of their works to train AI Claude. The $1.5 billion settlement could have been a record in copyright cases: about 500,000 authors and publishers expected to receive $3,000 for each work, Engadget reports.

Alsap called the agreement "far from complete" and criticized the lawyers for missing key details: a list of works and authors, a mechanism for notifying plaintiffs, and a form for filing claims. He warned that the agreement could be "imposed on the authors" and that in class-action lawsuits, the interests of victims often take a back seat once the amount of compensation is determined.

The lawsuit, Bartz v. Anthropic, was filed in 2024 by writer Andrea Bartz and nonfiction authors Charles Graeber and Kirk Wallace Johnson. They alleged that Anthropic had downloaded more than 7 million books from pirate libraries like LibGen and PiLiMi, creating a “central library” for training AI. In June, a judge issued a mixed verdict: he found that training on legally purchased books was transformative “fair use,” but obtaining works from pirated sources was a copyright violation.

The judge ordered the lawyers to submit a final list of works by September 15, and a list of participants and a form of claims that must pass the court's review by October 10. Only then will he consider preliminary approval of the agreement. The final version must also include guarantees that Anthropic will not be tried again for the same actions.

This case has become one of the most high-profile in a wave of lawsuits against AI companies, and its outcome could determine the approach to the use of copyrighted materials in AI training.