Anthropic Destroyed Millions of Books to Train Claude Legally

Artificial intelligence firm Anthropic bought millions of used books, sliced off their covers, and scanned every page to train its Claude model. A federal judge recently ruled that destroying physical books for this digital conversion is completely legal. Court files revealed the initiative under the internal name Project Panama, showing that the company eventually turned to purchasing print copies after starting with pirated digital libraries.

Judge William Alsup applied the first sale doctrine, which dictates that when you buy a book, you legally own it and can resell, lend, or bin it. The ruling noted that converting a purchased print copy into a digital PDF while destroying the original creates no extra copies. Because the physical book was destroyed to save storage space, the total number of copies remained at one, keeping the process within legal boundaries.

While the legal battle over physical copies was won, Anthropic still faces heavy costs for past piracy. The company recently faced a major payout for using unauthorized digital book databases and deals with separate copyright claims from music publishers. Meanwhile, the high demand for clean, human written text highlights the growing value of verified data provenance, a challenge that also drives interest in blockchain based data ownership solutions across the tech sector.