AI Firms Buy and Destroy Rare Books to Train Models

AI companies, including Amazon and Anthropic, are purchasing large quantities of rare and obscure books—such as home oxygen treatment guides, medieval English marriage litigation texts, and Swedish comedy volumes—to train their large language models. These books are often acquired from dealers at whatever price is asked, leading to a spike in income for sellers. The books are then sent to scanning facilities, such as one in Las Vegas run by Amazon, where they are digitized and subsequently destroyed. According to 404 Media reporter Emanuel Maiberg, the process involves cutting off the spines and feeding pages through scanners to speed up digitization. This practice addresses the problem of 'model collapse,' where AI trained on AI-generated data degrades in quality. The Wall Street Journal's Melissa Korn describes this as 'an unusually on-the-nose example of digital technology devouring literary culture.' While many of these books are technically rare, they are often esoteric items like vanity publications and old instruction manuals rather than first-edition classics. The trend highlights the lengths AI companies will go to secure human-written text, even at the cost of physical books.