Anthropic to pay $1.5B to authors in landmark AI settlement (www.theverge.com)

Anthropic has agreed to pay at least $1.5 billion, plus interest, to authors in a landmark class-action settlement over the company's use of copyrighted books to train its AI models. This payout, estimated at roughly $3,000 per work, is believed to be the largest ever in U.S. copyright litigation and marks a significant precedent for AI training data practices. The settlement covers around 500,000 works and requires Anthropic to destroy all original downloaded files, with additional payments possible depending on the number of claims submitted. Crucially, the agreement only covers past uses, explicitly excluding licensing for future AI training beyond August 2025. The case stems from allegations that Anthropic built its multimillion-dollar AI business by using pirated copyrighted material. Although a federal judge recently ruled that training AI on legally purchased books counts as fair use, the lawsuit advanced on Anthropic’s alleged use of pirated content. The settlement resolves these legacy claims without granting Anthropic ongoing rights, emphasizing the complexity of copyright issues in AI development. This comes amid rising legal scrutiny of AI companies over unauthorized data use, alongside growing corporate partnerships to license training data legitimately. Technically, this settlement underscores the increasing need for transparent, lawful sourcing of training data in AI and may prompt tighter regulatory and industry standards. For authors and rights holders, the case demonstrates a new mechanism for compensation tied directly to AI training and sets a precedent likely to influence future copyright debates and AI’s evolution. Anthropic's commitment to destroying infringing data and limiting the settlement’s scope signals cautious progress in balancing innovation with creators’ rights.