A federal judge in San Francisco approved Anthropic's $1.5 billion settlement Monday with a class of authors and publishers who sued the company for downloading their books from piracy websites to train Claude. U.S. District Judge Araceli Martinez-Olguin signed off on the deal over objections from some authors, calling their complaints "not grounded in a realistic assessment of the overall risks and rewards of a trial." Authors filed Bartz v. Anthropic in August 2024, alleging Anthropic had downloaded more than 7 million books from the shadow libraries Library Genesis and Pirate Library Mirror. A prior ruling already held that AI training on lawfully acquired books is fair use. That ruling stands undisturbed. The settlement covers roughly 482,000 registered works at about $3,000 each; more than 91% of eligible rightsholders filed claims.

The company paid for its piracy, the class got the largest copyright recovery in history, and the fair-use ruling means authors had no better hand to play at trial.

Class counsel say losing at trial would have paid the class nothing. Co-lead class counsel Justin Nelson of Susman Godfrey LLP called the outcome "the largest publicly known copyright recovery in history." His co-counsel Rachel Geman of Lieff Cabraser Heimann & Bernstein agreed: "The settlement is historic. It reminds us of the paramount importance of books." Judge Martinez-Olguin underlined that math in her approval order: "Success at trial was not assured, and a loss would have left the class with no recourse."

Anthropic says the settlement closes a narrow, bounded question. Aparna Sridhar, Anthropic's Deputy General Counsel, said in a statement: "We reached this settlement in 2025, after the court's landmark ruling that training AI on books is fair use under copyright law — which remains the law today." The company must destroy all copies of the pirated library within 30 days. Its training data from lawfully acquired books is untouched, and the fair-use ruling stands.

The Authors Guild CEO, Mary Rasenberger, called the outcome a necessary corrective. She pointed to the 91% claim rate — far above the typical 10% in class actions — as evidence the class found real value in the deal. She also noted that the companies that pirated millions of books to build AI "stole from those earning a median income of barely $20,000 a year." Anthropic must pay the $1.5 billion in four installments; the next $300 million is due by approximately July 28, 2026.

Authors Who Objected: $3,000 a Book Is Pocket Change

The settlement buys Anthropic out of the piracy claim but lets it keep profiting from the same books indefinitely.

Three thousand dollars per work is 2% of what a jury can award. Objector Robert Jacobson raised that comparison at the May 14 fairness hearing, noting the statutory ceiling is $150,000 per infringed work. Randy Hoopman put the frustration more plainly: "I don't really understand how it can be just if Anthropic is continuing to profit from essentially stealing our work even after the settlement is complete." Objectors' attorney James Bartolomei argued the settlement notices had been inadequate: class members never received clear information about the fair-use ruling their attorneys had already lost. He said that ruling was central to any informed decision about opting out.

The Authors Guild backed the settlement and said the per-book amount "feels paltry." The Guild was also blunt about the deal's limits: it covers only Anthropic's piracy, not AI training itself. "We strongly disagree with the training decision," the Guild wrote, referring to Judge Alsup's fair-use ruling that lets Anthropic train on lawfully acquired books without compensating authors at all. Catherynne M. Valente, a science fiction and fantasy author with 15 books on Anthropic's pirated list, called the payout "blood money" in a widely shared Substack essay. She wrote that she would take it because she needed the money, but that "$1.5 billion is only a lot to us. It's nothing to the forces behind Anthropic."

Authors Who Opted Out: Only a Jury Trial Gets Real Justice

The class settlement erased the piracy claim; these authors want a jury to price the infringement at what the law actually allows — and want Anthropic banned from using their work at all.

Twenty-eight authors rejected the settlement and filed their own suit against Anthropic. Novelists Dave Eggers, Angie Cruz, and Vendela Vida, Pulitzer Prize-winner Andrew Sean Greer, and memoirist Tobias Wolff are among the plaintiffs in Cruz v. Anthropic, filed May 14. The suit demands a jury trial on statutory damages — up to $150,000 per work — and a permanent injunction barring Anthropic from using their books in Claude training. A second group of 34 authors, in a parallel action called Kwon v. Anthropic, filed two days later with the same demands.

They believe a jury would price the piracy far above $3,000 per book. The opt-out plaintiffs are betting that a jury, rather than a mediated class deal, will set the price closer to what Congress allowed when it wrote the Copyright Act. They are also seeking something no class member can claim: a court order that stops Anthropic from building Claude on their work.

The opt-outs refused that trade: accepting $3,000 per book would have closed the piracy claim forever. The 28 authors in Cruz v. Anthropic believe the piracy was willful and a jury would price it far closer to the $150,000-per-work ceiling. Bloomberg Law has already called the $1.5 billion settlement "the bar for OpenAI, Meta," framing it as a floor rather than a ceiling for pending AI copyright cases.

Where This Lands

The settlement establishes one clear outcome: downloading pirated books to train AI is infringement, and Anthropic paid $1.5 billion to settle that claim. Whether AI companies owe authors anything for training on books they acquired legally stays open. Multiple legal analysts note Judge Alsup's fair-use ruling creates no binding appellate precedent. The 28 opt-outs in Cruz v. Anthropic will test whether a jury prices the piracy differently than a class negotiation did. Music publishers have already filed a separate $3.1 billion suit against Anthropic over pirated song lyrics. And cases against OpenAI, Meta, and Stability AI are all pending. What authors are owed for books that trained AI systems remains unanswered. The settlement didn't resolve it, and no appellate court has either.

Sources