A federal court in San Francisco has finally approved a settlement between Anthropic and the authors of books who accused the company of copyright infringement. The maker of the Claude chatbot will pay $1.5 billion into a fund for authors and publishers. The agreement is described as the largest known copyright settlement in the United States.
The case covers 482,460 books. The estimated compensation amounts to around $3,000 per title, before deduction of legal costs and lawyers’ fees. By mid-April, claims had been filed for 440,490 titles, representing 91.3 per cent of all works covered by the settlement. The court awarded the authors’ representatives approximately $101.6 million, significantly less than the $187.5 million requested.
However, the scope of the case is of key importance. In June 2025, Judge William Alsup ruled that the use of legally obtained books to train AI models may fall within the scope of the US principle of fair use. At the same time, he stated that downloading books from illegal online libraries and storing them in a central archive was not protected by this principle.
The settlement therefore primarily concerns the method of obtaining the material, rather than the process of training artificial intelligence itself. It covers books downloaded from the LibGen and Pirate Library Mirror websites. Anthropic is to delete these files, with the exception of copies it is required to retain for legal reasons. The agreement settles past claims and does not cover future works or content generated by AI models.
The decision provides the industry with its first concrete benchmark for assessing the financial risks associated with the use of pirated datasets. However, it does not provide a broader ruling on the terms under which technology companies may train models on legally obtained but copyright-protected content. The answer to this question will depend on future legal proceedings and rulings.
