Anthropic to pay $1.5B to settle authors' suit over AI training data

Anthropic has agreed to pay $1.5 billion to settle a class action brought by book authors who say the artificial intelligence company used pirated copies of their works to train its Claude chatbot, according to court filings disclosed Friday. The settlement, which must be approved by a judge, would pay roughly $3,000 for each of an estimated 500,000 class works and is being called the largest copyright recovery in U.S. history.

The agreement follows a June summary judgment by U.S. District Judge William Alsup that split the case. Alsup found that training AI on books obtained legitimately could qualify as fair use because it was highly transformative, but he also ruled that Anthropic had wrongfully acquired millions of books from pirate websites to build its training datasets. Plaintiffs had been preparing for a December trial on the question of liability for those downloads before the parties reached the settlement.

Plaintiffs attorneys said the payout will provide meaningful compensation to authors and publishers and could reshape how AI companies acquire and use creative works. Justin A. Nelson, a lawyer representing the authors, said the settlement sends a powerful message...