IP Law Daily, COPYRIGHT—N.D. Cal.: $1.5 billion settlement in Anthropic AI case gets preliminary approval, (Sep 26, 2025)

By Matthew Hersh, J.D.

The settlement, the largest in copyright history, will place a damper AI services’ used of pirated materials to train their models.

A federal judge in San Francisco has given his preliminary approval to a $1.5 billion settlement over the pirated use of books to train the large language model powering a popular AI “chatbox” service, according to public reports and a brief docket order filed by the court. The settlement, described by the parties as “the largest copyright settlement in the world,” comes shortly after the parties filed a joint response to 34 detailed questions that the court had posed in response to settlement agreement originally submitted—and widely reported on—earlier this month (Bartz v. Anthropic PBC, No. 3:24-cv-05417-WHA (N.D. Cal. Sept. 25, 2025)).

The settlement comes in the putative class action lawsuit filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson. The three authors took aim at Anthropic PBC, an AI software firm that runs the chatbox known as Claude. When a user prompts Claude with text, Claude responds with text that mimics human reading and writing. Claude can do so, the authors assert, because Anthropic trained Claude using books and other texts—including many of their own works—selected from a central library Anthropic had assembled. Anthropic assembled this library in two different ways: partly by downloading books from pirate websites, and partly by purchasing used print copies and scanning them into its database.

The lawsuit revolves around three different sets of allegedly unauthorized uses by Anthropic. First, the lawsuit alleges, Anthropic infringed upon the book authors’ copyrights by using their works to train the large language model underlying Claude. Second, the lawsuit alleges, Anthropic infringed by purchasing used print versions of their books and scanning them into digital form. Third, the lawsuit alleges, Anthropic infringed by downloading pirated versions of their books, training on some of them, and retaining all of them in their database for potential further uses.

Three events took place in rapid session over the summer that, almost certainly, led to this settlement. First, in late June, the court found that even though Anthropic’s ingestion of legally acquired books to train its model constituted a fair use, the AI company would be on the hook for the more than seven million books that it downloaded—and then retained in its files—from pirate sites Library Genesis and Pirate Library Mirror. (A person who downloads works from a pirate site “has infringed already, full stop,” the court emphasized.) Then, a few months later, the same court certified a class of all copyright owners of any of those seven million illicitly download books—threatening Anthropic with a multi-billion dollar tab in statutory damages. Finally, last month the court refused to stay the two decisions while the Ninth Circuit considered whether to accept an interlocutory appeal.

The terms of the settlement were first reported on September 5, when the parties filed a joint motion for approval of the settlement. The motion described the terms of the settlement as follows:

• “Anthropic will pay the Class at least $1.5 billion dollars, plus interest. With around 500,000 works in the Class, this amounts to an estimated gross recovery of $3,000 per Class Work.

• Anthropic will destroy the LibGen and PiLiMi datasets after the expiration of any litigation preservation or other court orders.

• In exchange, Anthropic will receive a past release only for conduct up to August 25, 2025. Claims arising out of conduct after August 25, 2025 won’t be released by the Settlement, nor will any claims (past or future) arising out of allegedly infringing outputs from Anthropic’s AI models.

The parties described the settlement, in their joint motion, as “nothing short of remarkable.” If approved, the parties stated in their filing, “this landmark settlement will be the largest publicly reported copyright recovery in history, larger than any other copyright class action settlement or any individual copyright case litigated to final judgment. The proposed settlement “well surpasses other copyright recoveries,” will provide “meaningful compensation” for each work covered by the class, the parties stated, and “will set a precedent of AI companies paying for their use of pirated websites like Library Genesis and Pirate Library Mirror.”

The settlement drew one step closer when the parties submitted a joint response to a set of 34 highly detailed questions—one set issued on September 10 and a follow-up set issued on September 16—from the court overseeing the would-be settlement. The first set of questions asked for the parties to address how the settlement would address 17 different “scenarios” set out by the court. (Examples: “Author and publisher file a joint claim providing adequate documentation of their beneficial and legal ownership of the right to reproduce copies and stating they have reached agreement on how to split their award”; “Author and publisher file joint (or separate) claims stating they have not reached agreement on how to split their award and refuse to submit the dispute to a special master,”) The second set of questions set out further scenarios envisioned by the court (examples: “If someone submits a claim, can they later opt out?”; “Some works on the Works List are not on the Class List. And, many listed on the Class List are missing addresses. How will this be cured, and cured timely?”).

The answers appeared to satisfy the court. After a 30-minute hearing yesterday, the following appeared on the docket: “The Court granted preliminary approval to the class settlement, preliminary approval to the plan of distribution, and preliminary approval to the plan for notice. Parties to comply with the Courts directives as stated on the record.” As of press time, no further order, other than brief docket order, had been filed.

The Case is No. 3:24-cv-05417-WHA.

Judge: Alsup, W.

Attorneys: Justin A. Nelson (Susman Godfrey L.L.P.) and Rachel Geman (Lieff Carbraser Heimann & Bernstein LLP) for Andrea Bartz, Andrea Bartz, Inc., Charles Graeber, and MJ + KJ Inc. Douglas A. Winthrop (Arnold & Porter Kaye Scholer LLP), Daralyn J. Durie (Morrison & Foerster LLP), Kathleen R. Hartnett (Cooley LLP), and Mark Alan Lemley (Lex Lumina LLP) for Anthropic PBC.

Companies: Andrea Bartz, Inc.; MJ + KJ Inc.; Anthropic PBC

MainStory: TopStory AINews Copyright TechnologyInternet CaliforniaNews GCNNews