Anthropic's Project Panama: Secretly Shredding Books for AI

Recently unsealed court documents reveal how Anthropic secretly bought, destroyed, and scanned books to train its AI chatbot, Claude. This covert operation was known as Project Panama. Anthropic's internal planning documents outlined a secret plan to "destructively scan all the books in the world," with the explicit instruction, "We don’t want it to be known that we are working on this.” Project Panama began in early 2024. Anthropic executives sought books not available online to teach Claude "how to write well," aiming to avoid "low quality internet speak." The startup purchased tens of thousands of books from vendors like Better World Books and World of Books. They then outsourced the scanning and destruction, using machines to slice book spines and scan pages. One vendor was asked to convert 500,000 to two million books over six months. Anthropic believed this initiative would lead to higher quality AI communication and address AI model collapse, which happens when AI trains on its own output. Books offer vast amounts of human-written, professionally edited literature, unlike much of the modern web. This revelation offers a clear look at how a major AI firm views the work of creators.