A federal judge has granted final approval to Anthropic’s $1.5 billion settlement with authors and publishers over books the company downloaded from pirate sites to train its Claude models, closing the largest copyright class-action settlement in US history.
Judge Araceli Martínez-Olguín of the US District Court for the Northern District of California signed the order on July 21, 2026, according to court filings. She took over the case, known as Bartz v. Anthropic, after Judge William Alsup — who ruled last year that training AI on lawfully acquired copyrighted text is fair use, but that downloading books from piracy sites like Library Genesis was not — retired.
What the settlement pays
The deal compensates rights holders for roughly 500,000 works at about $3,000 per title. Court filings cited a claims rate near 93%, and the order notes the per-work amount is roughly four times the statutory minimum for both willful and non-willful infringement. Martínez-Olguín’s order also set attorneys’ fees at approximately $101.5 million — well below the $187.5 million originally requested — and reduced service awards for the named plaintiffs to $15,000 each from the $50,000 sought.
Objections rejected, no binding precedent
The order records that some of the strongest objections came from people outside the settlement class, whose concerns the court said it could not weigh in evaluating fairness to class members, while noting they remain free to pursue their own claims separately. Because Anthropic settled instead of appealing, Alsup’s fair-use ruling never reached an appeals court, so it sets no binding precedent for the other AI copyright suits still pending against Google, Meta, OpenAI, and Midjourney.
The Association of American Publishers welcomed the outcome; its president, Maria Pallante, said sourcing training data from piracy sites “is not a choice we should simply accept as an efficiency for the infringer.” Under the settlement, Anthropic is also required to destroy the pirated files it originally downloaded. Payments to class members are set to begin roughly 30 days after the order becomes final, assuming no appeal is filed.
The ruling lands as Anthropic has separately been fighting distillation accusations against Alibaba, and as other publishers continue pressing AI companies over training data in court — including a bid by a group of newspapers to sanction OpenAI in a separate case.