They approve the agreement by which Anthropic will pay 1.5 billion dollars to train its AI with pirated books

Anthropic will finally pay $1.5 billion in copyright compensation for training its artificial intelligence (AI) models with pirated books after the deal was finally approved Monday.

The technology firm reached a settlement last year with a group of writers who sued it on the grounds that it had illegally used “hundreds of thousands of copyrighted books” to train its Claude models. Some of those books he had purchased while another came from a database of pirated content.

The agreement, which establishes the payment of 1.5 billion dollars (around 1.315 million euros at the exchange rate) plus interest, received preliminary approval at the end of September by federal judge William Alsup, despite his initial rejection and after calling the parties to answer some questions.

The reason given for rejecting it was that Anthropic could be imposing the agreement on the writers, since its acceptance would mean the closure of the case, preventing it from continuing through appeals and reaching a decision that would be binding on the entire industry.

Alsup partially ruled in Anthropic’s favor, declaring that the use of legally acquired books to train AI models amounts to a sufficiently “transformative” use. However, the judge also ordered a new trial to investigate the use of pirated books, questioning why the company chose to download them illegally rather than purchase them.

This Monday, and after Alsup’s retirement, Judge Araceli Martínez-Olguín finally approved the agreement, as reported in TechCrunch.

The agreement covers nearly 500,000 works, whose copyright is in the hands of writers and editors, with the payment of 3,000 dollars (about 2,627 euros) for each of them.