Overview
- A federal judge approved the $1.5 billion class settlement on Tuesday, resolving claims that Anthropic downloaded and retained millions of pirated books used during development of its Claude models.
- The agreement covers more than 480,000 works with roughly 91–92% of those titles claimed and anticipates payments near $3,000 per claimed work before legal and administrative deductions.
- Anthropic has agreed to destroy the identified pirated copies but did not admit broad liability for training its models, and some authors and publishers have opted out to pursue separate lawsuits.
- The court record and prior rulings drew a legal line between transformative fair use for model training and unlawful acquisition or retention of works from pirate libraries, a distinction likely to shape how companies source training data.
- The settlement avoids a December trial that carried large statutory damages exposure, attorney fees were authorized, and related copyright suits against other AI firms remain pending and consequential for the industry.