Anthropic's $1.5B copyright settlement stalls as authors and publishers fight over the payout

The $1.5 billion settlement Anthropic agreed to pay authors over its use of pirated books has turned into an internal fight, with writers and publishers at odds over how the money should be divided and who gets to decide. The dispute now sits before the court that oversaw the class action, where the two groups are pressing competing plans for distributing the largest copyright payout ever reached in the United States.

The settlement, announced after authors sued Anthropic over its practice of training Claude on books obtained from shadow libraries, covers hundreds of thousands of works. That scale is the problem: no one agrees on a formula for splitting the money between the people who wrote the books and the publishing houses that hold contracts on them. Authors argue the class is made up of writers, not corporations, while publishers say their agreements give them control over a large share of the covered titles.

The mechanics of the claims process are also in dispute. Both sides have proposed different methods for identifying eligible works and verifying ownership, and each wants a say in who administers the fund. Those choices matter beyond this case, because the procedures set here are likely to become a template for the other AI copyright suits still working through the courts.

Anthropic has not disputed the core allegations that it downloaded and stored pirated books, and the company agreed to the payout without admitting wrongdoing. For the AI industry, the number is the message: $1.5 billion is far beyond what any model developer has previously paid to settle training-data claims, and it puts a price tag on the risk of using scraped books. Rivals facing similar suits now have a benchmark for what a loss, or a deal, could cost.

For authors, the stakes are more immediate. Many of the writers whose books are covered have never seen a payment from a tech company for the use of their work, and the fight over who gets paid first could delay checks by months. Smaller authors and estates with limited resources are the most exposed if the process favors large publishers with legal teams already in the room.

What happens next rests with the judge, who will review the competing distribution proposals and could order mediation between the parties. The open question is whether the court imposes a single formula or lets authors and publishers negotiate a split themselves. Either way, the ruling will shape how future AI training-data settlements are structured, and how quickly the writers behind the books actually get paid.