Millions of physical books are being pulled into the race to train AI. Across the AI industry, companies are looking for high-quality, human-written text, and books are especially valuable because they contain edited, structured writing from before the internet became flooded with AI-generated content. Anthropic is one of the clearest examples. Its “Project Panama” bought millions of books, cut off their bindings, scanned the pages, and sent what remained for recycling. But destroying books isn’t necessary. Machines like the Treventus ScanRobot 2.0 can scan bound books without cutting off their spines, at up to 2,500 pages per hour. The reason companies use destructive scanning is simple: once the spine is removed, loose pages can be processed faster and more cheaply at industrial scale. Recent reports have also raised concerns about older, obscure, and potentially rare books being pulled into these AI data pipelines. AI needs massive amounts of human-created data. But if we ca...