Anthropic Faces $1.5 Billion Copyright Settlement for Unauthorized AI Training Data
Key Takeaways
- ▸$1.5 billion settlement expected to be approved for Anthropic's unauthorized use of copyrighted works in AI training
- ▸Only ~500,000 of 7 million pirated works qualify for payouts, with eligible authors receiving ~$3,100 per work
- ▸Settlement requires strict copyright registration criteria, excluding unregistered and many self-published works
Summary
Anthropic has reached a landmark $1.5 billion settlement in a class action lawsuit over its use of pirated books to train its large language models. The final Fairness Hearing on the settlement was recently held, with Judge Martinez-Olguin expected to approve the settlement, marking an unprecedented moment in copyright law enforcement against AI companies.
The settlement will distribute funds to authors and publishers whose copyrighted works were downloaded from pirate libraries like LibGen and PiLiMi without authorization. Eligible claimants—those who registered their copyrights with the U.S. Copyright Office—will share approximately $3,100 per pirated work. Authors whose works had reverted publishing rights or were self-published will receive the full amount without sharing.
However, the settlement's impact is more limited than headlines suggest. Of an estimated 7 million copyrighted works pirated by Anthropic, fewer than 500,000 qualify for the settlement based on strict registration requirements. As of the May 14 hearing, 447,576 works were claimed—roughly 7% of all pirated materials. The narrow class definition requires works to have ISBN/ASIN numbers and U.S. Copyright Office registration within five years of publication, effectively excluding significant portions of the literary community.
- Similar lawsuit against Meta signals broader legal liability for AI companies over training data sourcing practices
