Executive Summary
Anthropic, the AI developer behind the Claude language model, has agreed to a proposed class-action settlement totaling at least $1.5 billion, pending court approval. This landmark resolution—among the largest in U.S. copyright history—targets claimed unauthorized use of pirated books during model training, covering up to 500,000 works at roughly $3,000 per work. In addition, Anthropic will destroy the allegedly pirated datasets, and the settlement resolves claims relating to actions taken through August 25, 2025.
Discussion
On September 5, 2025, Anthropic, the Artificial Intelligence (“AI”) developer behind the chatbot “Claude,” agreed to a proposed class action settlement totaling at least $1.5 billion. If approved, the settlement would be one of the largest in the history of copyright law.
The Bartz v. Anthropic case began in August 2024 when authors Andrea Bartz, Charles Graber, and Kirk Wallace Johnson sued Anthropic, alleging the company used pirated copies of their copyrighted books to train Anthropic’s AI system without permission. Anthropic defended against these allegations, asserting its copying constituted permissible fair use under Section 107 of the Copyright Act.
In a June 2025 summary judgment ruling on Anthropic’s fair use defense, Judge William Alsup of the U.S. District Court for the Northern District of California held that while training AI systems on legally obtained, copyrighted content may constitute fair use, Anthropic’s practice of downloading and storing large quantities of books from purported piracy sites (e.g. Library Genesis and Pirate Library Mirror) should not receive protection. That ruling meant Anthropic would potentially face trial on the piracy claims. The ruling provided each side with a partial “win”: Anthropic prevailed on certain aspects of its fair use arguments, but the plaintiffs were allowed to proceed to trial as to Anthropic’s use of pirated works.
The proposed settlement covers approximately 500,000 works and averages about $3,000 per protected work. Anthropic will also have to destroy the infringing material. In exchange, Anthropic will receive a release of liability for conduct up to August 25, 2025. However, the release does not cover future claims relating to the “reproduction, distribution, and/or creation of derivative works.” Nor does it release Anthropic from future claims regarding Claude’s outputs, that is, its responses to user queries. Finally, the release only covers works on the list agreed to by all parties.
In an order issued in advance of a September 8, 2025, settlement hearing, Judge Alsup postponed approval of the settlement proposal, noting his disappointment that counsel on both sides had “left important questions to be answered in the future” that “need to be confirmed” before preliminary approval could be granted. These questions include the “Works List,” the processes for notification of claimants, allocation, and dispute resolution.
Following the preliminary settlement hearing, impacted rights holders should monitor the outcome, deadlines and claims submission protocols, and specifics around attorneys’ fees and settlement administration.
As the other major AI-copyright litigations wind their way through District Courts across the country, the contours of fair use will continue to come into focus. For now, the Anthropic settlement offers current and future litigants an interesting window into how at least these parties have opted to valuate activities arguably falling outside the fair use doctrine.
