Anthropic settles AI book piracy class action

Anthropic just dodged a trial that could have cost it up to $1 trillion.

The Amazon-backed AI startup reached a proposed class-action settlement with a group of authors who accused the company of training Claude on millions of pirated books. Filed in the Ninth Circuit Court of Appeals on Tuesday, the deal allows Anthropic to skip a high-stakes trial that was scheduled for December.

Fair use won't save you from Napster tactics

The lawsuit, originally filed last year by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, claimed Anthropic fed its Claude models an open-source dataset packed with pirated works.

Anthropic actually scored a major win back in June. Judge William Alsup ruled that training AI models on legally purchased books falls under fair use. At the time, Anthropic celebrated the decision, telling reporters that using books solely to build LLMs was officially fair play.

Then came July, and things got messy.

Judge Alsup allowed the authors to move forward with a class action, pointing out that Anthropic allegedly engaged in Napster-style downloading of millions of copyrighted works. Even with fair use protections on the actual training, raw digital piracy carries massive statutory damages. According to Wired, the startup was staring down billions, potentially over $1 trillion in penalties.

The financial terms remain confidential for now, but the settlement is expected to be finalized on September 3rd. Justin Nelson, an attorney for the authors, called it a historic settlement that will benefit all class members. Anthropic declined to comment.

Why it matters

Anthropic just showed the rest of the AI industry where the real legal minefield lies.

For the past two years, AI labs have leaned heavily on the argument that model training transforms text and qualifies as fair use. But this case proves that the front door still matters. Even if the court agrees that training itself is legal, how you acquired the training data can still sink you.

We think settling before December was the only rational move for Anthropic. Sitting in front of a jury trying to justify Napster-style downloads with a theoretical $1 trillion liability hanging over your head is an existential gamble no startup should take.

The catch is that this sets a clear precedent for creators. By paying to make this class action go away, Anthropic effectively signals that authors have real leverage when piracy is involved. Expect more copyright owners to double down on probing the origins of early training datasets.