Anthropic has agreed to pay $1.5 billion to settle a class action lawsuit brought by approximately half a million authors whose books the company downloaded from pirate libraries. The settlement, which marks the largest payout in US copyright history, comes after a federal judge ruled that while the pirating was illegal, training AI on legally-acquired copyrighted works is protected under fair use doctrine.
Key Points
- Anthropic pays $3,000 per book to 500,000 authors — largest copyright settlement ever
- Judge ruled training AI on legally-obtained copyrighted works is "fair use"
- Settlement only covers pirated downloads, not the actual AI training use
Federal Judge William Alsup delivered a split decision in June that distinguished between two separate issues: how AI companies acquire training data and what they're allowed to do with it. While he found Anthropic liable for downloading millions of books from Library Genesis and Pirate Library Mirror — illegal repositories of copyrighted material — he also ruled that if copyrighted works are obtained legally, using them to train AI falls under fair use protections.
"The training use was a fair use," Alsup wrote in his ruling. "The technology at issue was among the most transformative many of us will see in our lifetimes."
The settlement requires Anthropic to pay approximately $3,000 per work to an estimated 500,000 authors. The company has also agreed to delete the pirated materials it downloaded from shadow libraries. Anthropic states it did not use these pirated books in any publicly released AI models.
Court documents revealed internal discussions at Anthropic about content acquisition. CEO Dario Amodei referred to the process of legally acquiring books as a "legal/practice/business slog" in internal communications. The company had downloaded materials from Library Genesis and Pirate Library Mirror, sites known for hosting pirated content.
Similar practices have emerged at other AI companies. During a deposition, Anthropic founder Ben Mann testified that he downloaded the Library Genesis dataset while working at OpenAI in 2019. Court documents from a separate case show Meta employees discussing Library Genesis as "a data set we know to be pirated." One Meta researcher expressed concerns, writing that using pirated material "should be beyond our ethical threshold."
According to court records, Meta staff downloaded nearly 82TB of data from shadow libraries including Anna's Archive, Z-Library, and LibGen for use in training their LLaMA models.
The settlement doesn't establish legal precedent since the case didn't go to trial. The Authors Guild is coordinating notifications to affected writers, helping them register to receive their portion of the settlement. The $3,000 per-book payment represents compensation for the unauthorized downloading, not for the AI training use itself.
"This landmark settlement is the first of its kind in the AI era," said Justin Nelson, an attorney representing the authors. "It will provide meaningful compensation for each class work and sets a precedent requiring AI companies to pay copyright owners."
Anthropic Deputy General Counsel Aparna Sridhar described the agreement as resolving the plaintiffs' "legacy claims," adding that the company "remains committed to developing safe AI systems that help people and organizations extend their capabilities."
Following legal pressure, Anthropic later adopted a different approach to content acquisition. The company hired Tom Turvey, former head of the Google Books project, who purchased physical books in bulk and had them scanned to create digital copies for AI training. Judge Alsup ruled this method complied with fair use doctrine.
The settlement requires court approval. A trial date had been set for December on the piracy claims, but the agreement between both parties effectively closes this phase of the litigation.
The ruling's implications extend beyond this case. If courts continue to find that training AI on legally-obtained copyrighted works qualifies as fair use, it could establish a framework for how AI companies approach content acquisition going forward.
The Anthropic settlement represents a significant moment in the ongoing debate over AI training and copyright. With the company having raised $13 billion recently, the $1.5 billion settlement amounts to a fraction of its funding. For the AI industry, the case provides clarity on one aspect of copyright law — acquisition methods matter — while leaving the door open for training on legally-obtained materials.
For authors and creators, the settlement provides compensation for past infringement while raising questions about future use of their work in AI systems. As more companies develop AI models and courts continue to weigh in on fair use, the relationship between copyright holders and AI developers continues to evolve.