US Court Draws Line Between AI Training and Book Piracy

Generative AI developers rely on vast libraries of books to train large language models, raising a central copyright question: whether copying protected works for machine learning qualifies as fair use. In a lawsuit against Anthropic, the US District Court for the Northern District of California separated the purpose of training from the method of acquiring the books, finding the former highly transformative without giving companies immunity for obtaining pirated copies.

On June 23, 2025, US District Judge William Alsup ruled that Anthropic’s use of books to train Claude was fair use. However, the company still faced claims over more than 7 million books downloaded from sources including Library Genesis and Pirate Library Mirror and retained in a central library. Anthropic later agreed to pay $1.5 billion to settle claims involving roughly 500,000 works, equivalent to about $3,000 per book.