anthropic / copyright settlement / ai training

A judge has delayed approval of Anthropic’s proposed $1.5 billion copyright settlement with authors and publishers over training AI models, introducing new procedural scrutiny and potential changes to the deal. The case centers on claims that Anthropic used copyrighted texts to train its models; the company agreed to the large settlement to avoid prolonged litigation but now faces renewed judicial review that could alter payments, injunctive terms, or class certification. Key players include Anthropic, plaintiffs representing authors and publishers, and the court overseeing the class-action settlement. The delay matters because it could reshape precedent around dataset use, model training practices, and liability exposure for AI developers across the industry.

A federal judge has delayed final approval of Anthropic’s proposed $1.5 billion copyright settlement over alleged book-piracy training of its AI, after multiple authors and class members objected to the deal. Objectors say lawyers are seeking excessive fees—more than $320 million—while individual authors would receive about $3,000 each, calling the payouts a “pittance.” Some argue attorney compensation was tied to the full fund rather than actual claimants and want fee reductions to boost author payments; one suggested cutting counsel fees to $70 million to raise plaintiff awards. The objections persuaded Judge Araceli Martinez-Olguin to ask for more explanation and coincided with some members opting out and filing a new suit.