Anthropic Consents To $1.5B Settlement Over Pirated Books

A lawsuit against Anthropic breaks new legal ground with $1.5 billion settlement for using pirated books to train its AI models.

Anthropic has reached a $1.5 billion preliminary settlement in Bartz v. Anthropic, which alleges that the company illegally trained its AI model, Claude, on a trove of pirated books. If greenlit, this agreement could be the most significant copyright recovery connected to AI training data in U.S. history.

Background and Legal Context

In June, Judge William Alsup ruled that training an AI on copyrighted books acquired legally was fair use, but copying millions of popular pirated books from sites such as Library Genesis and Pirate Library Mirror into its databases was infringement. That decision led to the current settlement talks.

Under the settlement, each eligible title would receive approximately $3,000, with a class size estimated at around 500,000 titles. Consumers claim Anthropic illegally downloaded a minimum of 7 million copies from piracy websites.

Counsel for the authors, Justin Nelson, said:

“As best as we can tell, it’s the largest copyright recovery ever.”

Payment Structure and Claims Process

As to the distribution of the settlement fund, the Authors Guild stated that it will be distributed in four payments after obtaining court approval:

• $300 million, a few months after initial approval
• $300 million after final approval
• Final approval, $12, $450 million
• $450M at 24 months following final approval

During this period, the funds that are held in escrow will also accrue interest. By October 10, a final “Works List” is due, which will steer a searchable database, helping rightsholders to submit claims. It prohibits the existence of pirated copies and only applies to past conduct.

AI-enabled tools are creating content, such as text and imagery, that blurs the line between original and derivative works, making content provenance more critical for creators and publishers than ever.

The settlement highlights changing views on transparency in licensing and training data. The payout per title sets a clear standard, which could bolster negotiating power in future licensing discussions between content holders and AI creators.

Preliminary Approval Review Outcome

Today, the Judge will review preliminary approval. If approved, a notice program for claimants will begin this fall.

Payments will be made according to the scheduled installments, following final court approval and verification of claims.

Bottom Line

As the use of AI in creative workflows remains in flux, this landmark settlement establishes a new precedent for AI copyright law, emphasizing the importance for innovative companies to source their data legally.