Full story
Settlement Approved in San Francisco
A federal judge approved a $1.5 billion copyright settlement in the class action Bartz v. Anthropic, resolving claims that Anthropic used pirated books to train its Claude chatbot.
“Federal Judge Araceli Martínez-Olguín granted, on July 20, 2026, final approval to the settlement reached in Bartz v”
The United States District Court for the Northern District of California validated the negotiated agreement on July 20, and the action was dismissed with prejudice as to the claims covered by the settlement.
The settlement requires Anthropic to pay $1.5 billion to the rights holders of 482,460 books recovered from LibGen and Pirate Library Mirror, and to destroy the files in question.
ActuaLitté reported that the court deemed the compromise “fair, reasonable and adequate,” and that the targeted versions were those retrieved from Library Genesis in June 2021 and Pirate Library Mirror in July 2022.
The AP described the ruling as providing “meaningful relief” to affected authors and publishers, with about 91% of the more than 482,000 books covered already claimed for payment.
Fair Use, Piracy, and Opt-Outs
The dispute hinged on a split between training and the acquisition of the underlying works, after Judge William Alsup found training AI chatbots on copyrighted books wasn’t illegal but that Anthropic wrongfully acquired millions of books through pirate websites.
ActuaLitté said Alsup held that using the books to train the models fell under fair use, but refused to extend that protection to the durable construction of a library of pirated copies.
In her order, Judge Araceli Martínez-Olguín overruled objections and emphasized that about 95 percent of the class received notifications and approximately 91 percent of authors and publishers impacted have already filed claims, while only 350 class members opted out.
Ars Technica reported that Martínez-Olguín noted the $3,000-per-work payouts were “four times the minimum statutory damages,” and that she cut lawyers’ fees down to less than 7 percent of the settlement fund.
The AP quoted Anthropic’s deputy general counsel Aparna Sridhar saying training AI on books is fair use under copyright law, with Sridhar highlighting the ruling Friday as a landmark showing “that training AI on books is fair use under copyright law.”
Payments, Destruction, and Next Steps
The settlement’s payment structure includes $300 million already paid in October 2025, $300 million after final approval, and two installments of $450 million in September 2026 and September 2027, with the latter two bearing interest.
ActuaLitté reported that Anthropic will have to delete the original files downloaded from LibGen and PiLiMi, as well as the copies that come from them, subject to its legal preservation obligations.
Courthouse News said the final approval granted final approval Monday with $1.5 billion awarded to authors and creators whose work was used by Anthropic to train its large language models, and that the settlement provides about $3,100 per work from approximately 482,000 works.
Courthouse News also stated that Anthropic will be required to destroy the original files of the works they downloaded from the pirated book datasets within 30 days of the final judgment and certify Library Genesis and Pirate Library Mirror datasets with pirated material were used in training any of the company’s commercially released large language models.
Plaintiffs’ attorney Justin Nelson said in an email to Courthouse News, “It is the largest known copyright recovery in history,” as the parties moved from a trial set for December 2025 to a settlement that closes the first major AI-related copyright dispute to reach resolution in the United States.
%2Fhttps%3A%2F%2Fi.s3.glbimg.com%2Fv1%2FAUTH_63b422c2caee4269b8b34177e8876b93%2Finternal_photos%2Fbs%2F2026%2FY%2FQ%2FvThFmBTwSlL6QYgbK7UQ%2Ffoto21bra-301-queima-a2.jpg&w=3840&q=75)
