A federal judge has granted final approval to Anthropic’s $1.5 billion class-action settlement with authors who alleged that the AI company downloaded pirated copies of copyrighted books for use in developing its models.

Judge Araceli Martínez-Olguín said in an order issued Monday that the agreement would provide meaningful relief to copyright holders. Eligible authors and publishers are expected to receive about $3,000 for each covered book, although the final amount may depend on the administration of claims and the settlement’s terms.

The case was filed in 2024 by authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson. It became one of the most closely watched legal disputes over the vast quantities of text used to train generative AI systems. The plaintiffs’ law firm has described the agreement as the largest known copyright recovery.

Anthropic agreed to the settlement in September 2025, and the court subsequently gave it preliminary approval. Final approval clears a major legal hurdle for distributing the fund, though the process of resolving claims and any remaining challenges may continue.

More than 91% of the authors and publishers covered by the agreement have claimed a share of the payment, Anthropic general counsel Aparna Sridhar told Reuters. She said the company looked forward to bringing the matter to a close.

The litigation drew an important distinction between using lawfully obtained books to train an AI model and acquiring those books from unauthorized online collections. In an earlier ruling, then-Judge William Alsup handed Anthropic a partial victory on the training question, finding that its use of books in creating its models could qualify as fair use. But that ruling did not shield the company from allegations involving the way it obtained and retained copies of the works.

The remaining class claim focused on what the court characterized as large-scale downloading of books from pirate libraries. That acquisition issue created substantial potential liability separate from the legality of analyzing books during model training. Anthropic chose to settle that portion of the dispute rather than proceed toward a trial over damages.

The distinction matters across the wider field of AI copyright litigation. Developers typically argue that training systems on existing material is transformative because models learn statistical patterns rather than simply republishing the underlying works. Authors, publishers and other rights holders have countered that companies should not be able to build commercial products from protected material without permission or compensation. Even where training may be protected as fair use, obtaining source material through piracy can present a separate and more direct infringement question.

The settlement does not resolve all copyright claims involving Anthropic. The company continues to face separate litigation, including a case involving Chicken Soup for the Soul and claims from other authors. Some challengers have also argued that compensation of roughly $3,000 per title is inadequate.

Nor does the approval establish a universal licensing rate for books used in AI development. It resolves the claims covered by this particular class action and reflects the specific allegations concerning Anthropic’s acquisition of the works. Other pending cases involve different datasets, methods of access and types of copyrighted content, leaving courts to address those disputes on their own records.

For members of the approved class, the next phase centers on settlement administration and payment. For Anthropic, final approval removes one major case from a growing docket of disputes over how AI companies source and use copyrighted material.

Sources: AI copyright lawsuit, AI copyright lawsuit, Anthropic