Anthropic settled a $1.5B copyright lawsuit — the price of training AI on pirated books

Anthropic agreed to a $1.5 billion class action settlement after being sued for downloading and storing millions of copyrighted books from pirate sites — Library Genesis and Pirate Library Mirror — to train its AI models. Federal Judge Araceli Martínez-Olguín granted final approval on July 20, 2026, finding the settlement provided 'substantial benefits' of approximately $3,000 per work across an estimated 500,000 works, four times the statutory minimum.

The settlement covers past conduct by Anthropic in acquiring and copying authors' works through August 25, 2025. Notably, Judge William Alsup had previously ruled that training AI on copyrighted text constitutes fair use, but the act of downloading from pirate sites was itself illegal — and that distinction forced Anthropic to settle rather than risk a trial. The court ordered Anthropic to destroy the pirated files. The settlement does not cover claims based on AI-generated outputs, and parallel copyright lawsuits against Google, Meta, Midjourney, and OpenAI remain ongoing.

Anthropic settled for $1.5B, paying $3,000 per work across 500,000 books, because downloading from pirate sites was theft, not fair use, and files were destroyed.

Anthropic's engineers chose convenience over legality — downloading from pirate sites rather than licensing content — and created a $1.5B liability that the fair-use defence could not protect

The settlement was a bet: pay $1.5B now to avoid a trial that could have established damaging precedent for the entire AI industry

The cost of cutting corners on training data is not a legal risk — it is a liability that compounds every day until the class action lands, and fair use does not cover theft