Anthropic agrees to a $1.5 billion copyright settlement with authors

Background

Bartz v. Anthropic was a class action filed by a group of authors against Anthropic. It was filed on 19 August 2024 in the Northern District of California ^[1]. The case concerned the books Anthropic used as training material for its AI models, and in particular the digital copies it made and collected along the way.

A class action is a lawsuit brought by a small number of people on behalf of a much larger group in the same situation. That structure matters here: a settlement in a class action binds the whole group, not just the named plaintiffs, which is why the scope of the release became one of the most closely watched parts of the deal.

The ruling that came first

Before any settlement, the court addressed the core copyright question on summary judgment. On 23 June 2025 it ruled that training on digital copies of books was fair use, but it ordered a trial over pirated copies ^[1][3]. Fair use is the doctrine that allows limited use of copyrighted material without permission, judged on factors such as the purpose of the use and the effect on the market for the original.

Publishers Weekly reported on 25 June 2025 that Judge William Alsup found the actual training process constituted fair use under copyright law ^[3]. The same ruling drew a line at how the copies were obtained: downloading pirated books from Books3, Library Genesis and Pirate Library Mirror to build a permanent digital library was not covered by fair use ^[3]. In other words, the court separated the act of training from the way the training data had been assembled, and only the first of those was protected.

The settlement

In September 2025 Anthropic agreed to pay $1.5 billion to settle the case ^[1]. Wikipedia describes it as the largest known US copyright settlement ^[1].

Judge Araceli Martinez-Olguin approved the settlement in July 2026 ^[1][2], and the court granted final approval on 20 July 2026 ^[2]. Final approval is the step that turns a negotiated agreement into a binding order covering the class.

The terms

Class counsel fees were set at about 6.8% of the settlement fund, or approximately $101.56 million ^[2]. The judge also cut the requested service awards for the three class representatives from $50,000 to $15,000 each ^[2]. Service awards are the extra payments given to the individuals who lent their names to the case and took on work on behalf of the class.

The release itself is narrow. It covers only claims about Anthropic's past acquisition and copying of the authors' works, through 25 August 2025 ^[2]. Claims based on AI outputs are not released under the settlement ^[2]. A release is the part of a settlement in which plaintiffs give up the right to sue again over the claims it covers, so its boundaries decide what remains open.

Why it matters

Two things happened in this case at almost the same time: a ruling that the training process itself was fair use, and a payment of $1.5 billion over how the books used in that training had been obtained. Read together, they show where the court placed the legal risk — not in learning from books, but in the source of the copies used to do it. The ruling on the pirated sources, and the price Anthropic agreed to pay to close off the claims attached to them, are the parts of the case most likely to be cited in later disputes.

The settlement's limits are equally significant. Because it reaches only past acquisition and copying through 25 August 2025, and because claims based on AI outputs are not released ^[2], the agreement did not resolve questions about what models produce. Those questions remain outside the deal, and the case therefore settled one part of the dispute while leaving another untouched.