Court Approves Record Anthropic Settlement: $1.5 Billion to Book Authors for Pirated Copies in Training Datasets

A US federal court has given final approval to Anthropic's $1.5 billion settlement with book authors, TechCrunch reported on July 20, 2026. It is the largest payout in a copyright case in US history and the first settlement of this scale between an AI company and rights holders.

What Anthropic is paying $1.5 billion for

Anthropic is paying for downloading pirated copies of books from shadow libraries and storing them in an internal dataset used to train its Claude models. The class action Bartz v. Anthropic was filed in 2024 by writers Andrea Bartz, Charles Graeber and Kirk Wallace Johnson. They showed that the company had downloaded millions of books from sources such as LibGen without the authors' permission.

• The settlement totals $1.5 billion, roughly $3,000 per work
• The compensation covers about 500,000 books
• Judge William Alsup gave preliminary approval in September 2025
• Final approval came in July 2026
• Anthropic has also committed to destroying the pirated copies in its storage

Payouts will be distributed through a dedicated fund: authors and publishers have been submitting claims since fall 2025. Some rights holders objected to the terms, arguing that $3,000 does not reflect a book's value to a commercial model, but the court found the settlement fair and reasonable.

Why the dispute over AI training is not settled

The settlement closes one specific case but does not answer the industry's central question: is it legal to train models on copyrighted texts? Back in June 2025, Judge Alsup ruled that training on legally purchased books is fair use, while downloading pirated copies is infringement. The $1.5 billion payment relates specifically to piracy. Had Anthropic lost at trial, statutory damages could have reached hundreds of billions of dollars, since US law allows up to $150,000 per willful infringement.

"The final approval closes one case, but it does not resolve the broader issue of using copyrighted works to train AI models,"

TechCrunch writes.

What this means for other AI companies

Other AI developers face dozens of similar lawsuits. By industry counts, more than 40 copyright cases against AI companies, including OpenAI and Meta, are pending in US courts. The most prominent is The New York Times' lawsuit against OpenAI and Microsoft, filed in December 2023 and still unresolved. In a parallel case against Meta, Judge Vince Chhabria sided with the company on fair use in June 2025, but stressed that the ruling applies only to the specific plaintiffs. Rights holders now have an argument: even the richest AI labs are willing to pay billions to keep a pirated-dataset case from reaching a jury.

What it means

The era of free training data for AI is ending. Training on legal copies is protected as fair use, but pirated datasets will cost billions. The market is already adjusting: publishers and media companies are signing licensing deals with OpenAI and Google, and the price of legal training data is rising.

Frequently asked questions

How much will each author receive?

Roughly $3,000 per book before legal costs are deducted. The $1.5 billion fund will be split among the rights holders of about 500,000 works that ended up in Anthropic's pirated dataset.

Did the court rule that training AI on books is illegal?

No. Back in June 2025, Judge William Alsup ruled that training models on legally acquired books is fair use. Anthropic is paying for downloading and storing pirated copies, not for the act of training itself.

*Meta has been designated an extremist organization and is banned in Russia.