A proposed $1.5B class action settlement has been filed against Anthropic, the company behind Claude AI. Authors allege that Anthropic used pirated books from shadow libraries (LibGen, Pirate Library Mirror, Books3) to train its models. About 465,000 works are on the preliminary list, with estimated payments of $3,000 per book if approved. The court has asked for a finalized Works List and sample claim form by late September. If granted, this would be one of the largest copyright-related settlements in U.S. history.

Beyond the payout, the case raises bigger questions: how should copyright law apply to generative AI? Can datasets built on unlicensed works be “cleansed”? And what standards should courts use when balancing innovation with IP rights?