The record copyright payout resolves claims over pirated books Anthropic stored, not the AI training a judge had already found to be fair use.
A federal judge in San Francisco, Araceli Martinez-Olguin, gave final approval on Monday to Anthropic's $1.5 billion settlement with a group of authors, closing a class action over the books used to train its Claude chatbot.
The payout stands as the largest ever recorded in a U.S. copyright dispute. It is also the first major case among the many dozens of AI-training suits — brought by authors and news organizations against tech firms — to reach a settlement.
The authors sued Anthropic in 2024, alleging the company, which counts Amazon and Alphabet among its backers, had used illegally copied editions of their books without permission to teach Claude how to answer people's prompts.
The money does not turn on the training itself. A court had already ruled that Anthropic's use of the authors' works to train Claude qualified as fair use. The settlement instead flows from a separate finding — that the company saved more than seven million pirated books into an internal repository, one not necessarily destined for AI training.
A trial had been scheduled for December to determine what Anthropic owed over the piracy it was accused of, with the company potentially exposed to as much as hundreds of billions of dollars.
In signing off, Martinez-Olguin turned aside objections that the sum was too small, cut the fee awarded to the plaintiffs' attorneys, and noted that rights holders for more than 92% of the 480,000 works covered had filed claims for payment.
Yet the record payout sets no legal precedent. The fair-use question had already been decided the previous year within the same suit, and the two district-court decisions to date have rested on narrow, case-specific facts — leaving the broader legal uncertainty over how AI models are trained unresolved.