For AI companies, books are often treated as raw training material. They may use digital copies, including allegedly pirated ones, as in cases involving Meta and Anthropic, which paid authors a $1.5 billion settlement. Some AI companies, including Anthropic, turned to cheap secondhand books when digital copies were unavailable. According to the lawsuit, Anthropic bought large quantities from resellers, used a hydraulic cutter to remove their bindings, scanned the loose pages with industrial equipment, and added the text to its AI training data, destroying the books in the process. The process relied on the first-sale doctrine, which lets buyers alter items they legally purchase. A judge also ruled that converting the books into digital training data was transformative and therefore protected as fair use. Originally built to help libraries, distributors, and bookstores locate and sell titles, ISBNdb now reportedly helps AI companies purchase between 1,000 and one million books at a time. (Source: 404 Media)