Anthropic Pays $1.5 Billion: Highest Copyright Settlement of All Time
An important milestone has been reached in the copyright dispute between AI giant Anthropic and a group of plaintiff authors, reports heise. US federal judge Araceli Martínez-Olguín has approved the out-of-court settlement between the Claude developer and the numerous authors in San Francisco.
The company is paying $1.5 billion for copyright infringements in the training of its AI models. Each affected work earns its author around $3,000. That is four times the minimum penalty for willful infringement.
Lawsuit over a dataset of “internet books”
More than 91 percent of those affected have filed their claims, according to Anthropic’s statement. The law firm in charge describes the sum as “the highest settlement amount ever awarded in a copyright case.”
The lawsuit was filed by three authors in August 2024; they later significantly expanded the proceedings. A paper by Anthropic employees from December 2021 served as evidence, revealing that 32 percent of a massive training dataset consists of “internet books” — a code word for books illegally downloaded from the internet. Anthropic put forward the billion-dollar settlement proposal in September; the plaintiffs accepted it immediately.
Anthropic Avoids Astronomical Penalties
The presiding judge did find that Anthropic’s use of copyrighted texts can in principle be covered by the fair use doctrine, since something new is created. However, the downloading of vast libraries of works compiled in violation of copyright did not fall under this. The court also established: Anthropic knew that the databases had been compiled illegally.
This finding put the Claude developer under massive pressure. In a trial, penalties of up to $150,000 per book could have been imposed. With hundreds of thousands of affected works, the total sum would have been astronomical — significantly higher than the $1.5 billion now agreed upon.
The case is part of a series of legal proceedings scrutinizing the practices used in training AI models. For Anthropic, the case ends with a record penalty — one that is still cheaper than the maximum risk of a lost trial.
