Overview

• The confidential settlement follows class certification in the Northern District of California, resolving authors’ claims that Anthropic built a permanent training corpus from millions of pirated books.
• Judge William Alsup previously ruled that AI training can be transformative fair use but found Anthropic’s acquisition and storage of books from sites like LibGen to be infringing.
• Class status opened the door to statutory damages up to $150,000 per work, with reported exposure reaching into the hundreds of billions and potentially over $1 trillion.
• The agreement averts a jury trial in this case, with terms undisclosed pending the court’s review and anticipated finalization in early September.
• Anthropic still faces separate copyright suits from music publishers and Reddit, reinforcing pressure on AI firms to license data and track provenance.