New York — Sept. 5, 2025: Anthropic agreed to a $1.5 billion settlement in a class-action brought by authors who say the company used pirated copies of their books to train its Claude chatbot; the deal, subject to a judge’s review, would offer roughly $3,000 for each of about 500,000 covered works and requires the company to delete the original pirated files.
Key Takeaways
- Anthropic will pay $1.5 billion to resolve a class-action by authors over pirated books used to train its AI.
- The settlement would provide roughly $3,000 per covered work for an estimated 500,000 books.
- A federal judge earlier found Anthropic downloaded more than 7 million digitized books from known pirate libraries.
- Anthropic has agreed to destroy the original book files it obtained through those sites.
- The case could influence parallel lawsuits against OpenAI, Microsoft, Meta, Midjourney and others.
- Some international authors may not be covered if their works lack U.S. copyright registration.
Verified Facts
A group of authors led by Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued Anthropic after court filings showed the company had obtained large numbers of digitized books from pirate sources. U.S. District Judge William Alsup ruled in June that while training AI on copyrighted works can qualify as fair use, Anthropic had wrongfully acquired millions of copies by downloading from known illegal sites.
Court filings and the judge’s summary identified collections used by Anthropic: about 200,000 books from a dataset called Books3, at least 5 million titles from Library Genesis (LibGen), and at least 2 million from the Pirate Library Mirror, a total the judge described as more than 7 million pirated books. The settlement narrows the pool for payment after removing duplicates and works outside the claim.
Under the proposed deal, authors or publishers would receive roughly $3,000 per eligible work for about 500,000 titles, producing the $1.5 billion total. As part of the agreement Anthropic has pledged to destroy the original files it downloaded from pirate libraries.
Anthropic, a private company founded in 2021 by former OpenAI employees, recently disclosed a valuation update to about $183 billion after new fundraising and projects projected annual sales in the billions, though the firm has not reported a profit.
Context & Impact
The settlement arrives amid a cluster of legal challenges over how tech firms obtain and use copyrighted material to train large language models. The case is closely watched because it separates the legal question of training as fair use from the separate wrongdoing of acquiring materials through piracy.
If approved, the deal could set a practical template for how creators and AI companies resolve mass-infringement claims without prolonged trials. It may also prompt firms to tighten dataset sourcing, invest in licensed corpora, or seek new licensing agreements with publishers and rights holders.
Advocacy groups and industry observers responded differently: authors groups hailed a meaningful recovery, while some rights defenders warned the settlement may be seen by tech companies as a cost of doing business rather than a full deterrent. European and other international creators may remain vulnerable if their works lack U.S. copyright registrations required for class claims.
Official Statements
Anthropic said resolving legacy claims is a step toward focusing on safe, useful AI development that benefits users and organizations.
Aparna Sridhar, Anthropic deputy general counsel
The Authors Guild described the agreement as a strong outcome that sends a message that pirating authors’ works to build AI carries consequences.
Mary Rasenberger, Authors Guild CEO
Unconfirmed
- How many non-U.S. authors will be excluded because of registration or jurisdictional limits remains uncertain.
- The precise impact on other pending suits against OpenAI, Microsoft, Meta and others depends on future filings and potential individual settlements.
Bottom Line
The proposed $1.5 billion settlement marks one of the largest recoveries tied to AI training datasets and could reshape how companies assemble training material. It underscores that sourcing practices matter: even if training can be lawful, acquiring copyrighted works through pirated channels carries legal and financial risk.
Approval by the court would resolve this chapter for Anthropic but is likely to accelerate licensing talks, policy scrutiny and further litigation across the AI industry.