The settlement pays for how Anthropic got the books, not for the AI training, after a 2024 ruling in the same case said training on books can be fair use.
A federal judge in San Francisco approved a roughly $1.5 billion class-action settlement on Monday. The dollar figure settles one piece of the case and leaves another intact: a 2024 ruling in the same lawsuit found that training an AI on copyrighted books can be fair use, and that holding still stands.
The settlement covers more than 482,000 books, and authors and publishers have already claimed about 91% of them. Anthropic will pay roughly $3,000 per book, a per-work figure the AP called an approximation; the actual distribution depends on the 91% claim pool. Plaintiff attorney Justin Nelson called the deal "the largest known copyright recovery in history" and said payouts would move as promptly as possible (AP News).
The court treated acquisition and training as two different products with two different price tags. U.S. District Judge William Alsup, who gave preliminary approval in September and has since retired, held in 2024 that training an AI on lawfully acquired books can be fair use, and that Anthropic's separate decision to pull millions of books from pirate libraries was not. The $1.5 billion pays for the download channel, not the training run. Anthropic's deputy general counsel Aparna Sridhar pointed to the fair-use portion of the earlier ruling as a landmark showing "that training AI on books is fair use under copyright law" (AP News).
Bestselling thriller novelist Andrea Bartz and two other authors filed the original suit in 2024. The case is now the first of dozens of AI copyright claims against frontier model labs to reach a settlement of this size. U.S. District Judge Araceli Martínez-Olguín signed off on the final approval Monday in San Francisco federal court.
A live question for the rest of the docket: when the next case involves a lab that bought its training corpus through licensing deals rather than pirate sites, will the fair-use defense the Alsup ruling preserved actually matter, or will it remain a procedural footnote that the Anthropic settlement bought without testing? Discussion on Hacker News has split along those lines (Hacker News thread).