Amazon AI facility secretly ordered 1,000 rare books for scanning

The investigation by 404 Media provides concrete evidence of a practice that has long been suspected in AI development circles: the covert bulk acquisition of books by major tech companies for training data. The use of an anonymous customer placing a very large order on Biblio—a strategy suggesting price insensitivity and deliberate opacity—mirrors the patterns previously documented when Anthropic's book-scanning efforts came to light in June 2025. The facility's entrance logo, a dinosaur clutching a book, suggests Amazon is not attempting to hide the operation's purpose from its own workforce, even if it remains opaque to the public and to booksellers. This discovery points to a systematic approach across the industry: companies identify marketplace channels where individual sellers operate with less scrutiny than major distributors, place enormous orders through apparently untraceable accounts, and transport the physical books to dedicated facilities for destructive scanning. The involvement of Amazon—one of the world's largest cloud and AI companies—indicates the practice extends far beyond a single player and may represent a standard acquisition method for training data in the competitive race to build ever-larger language models.