Anthropic Agrees to $1.5 Billion Settlement in Lawsuit Over Pirated Books Used to Train AI

Anthropic has agreed to pay $1.5 billion to settle a class-action lawsuit brought by authors who alleged the artificial intelligence company used pirated copies of their books to train its chatbot, Claude.

Under the proposed deal, which the parties have asked a federal judge to approve as soon as Monday, Anthropic would pay about $3,000 for each of an estimated 500,000 books covered by the settlement. The agreement also calls for Anthropic to destroy downloaded copies of books the authors said the company acquired improperly.

The suit was filed last year by three authors — thriller novelist Andrea Bartz and nonfiction writers Charles Graeber and Kirk Wallace Johnson — who later came to represent a broader class of writers and publishers. The plaintiffs said Anthropic downloaded millions of books from pirate websites to use as training data for Claude and alleged copyright infringement. In June, U.S. District Judge William Alsup issued a mixed ruling, concluding that training AI models on copyrighted books was not itself illegal but that Anthropic had wrongfully acquired books from pirate sites.

Plaintiffs said the proposed settlement, if approved, would be the largest publicly reported copyright recovery in history. "As best as we can tell, it’s the largest copyright recovery ever," Justin Nelson, a lawyer for the authors, said in filings. The settlement was announced by the parties in August without disclosing terms and was filed with the court this week to seek final approval.

Anthropic has backing from Amazon and is among several technology companies facing litigation over how they obtain and use copyrighted material to train generative AI systems. The Anthropic agreement marks the first major settlement in a wave of suits that have targeted OpenAI, Microsoft and Meta Platforms, among others.

In its court filing, the company and the plaintiffs asked Alsup to approve the terms and implement mechanisms for distribution of payments to authors and publishers. The settlement reportedly covers both authors and publishers whose works were identified as part of the materials Anthropic downloaded. The estimated $3,000 per book figure reflects the $1.5 billion total divided across the approximately 500,000 titles named in the agreement.

Legal experts and rights holders said the settlement could influence how AI companies source training data and how courts address the intersection of copyright law and machine learning. The June ruling that training itself did not constitute copyright infringement left open legal questions about acquisition methods and the proper remedies when protected works are used without authorization.

Anthropic said it would destroy the downloaded copies identified in the lawsuit. The company has not specified in the filings whether it will change broader data-acquisition practices or whether the settlement resolves every potential claim related to the books covered by the suit.

If Alsup approves the settlement, the resolution will be followed closely by authors, publishers and AI developers. The agreement could shape negotiations and litigation strategies in related cases alleging unauthorized use of creative works to train large language models and other generative systems.

The parties did not disclose additional operational terms in the court filing, and the judge will consider the fairness and scope of the settlement before deciding whether to grant final approval. The case has underscored tensions between content creators seeking compensation and attribution and technology firms seeking broad datasets to develop advanced AI models.