Investigation: AI firms suspected of buying and destroying rare books for training data

Book dealers are raising concerns that artificial intelligence companies may be systematically purchasing and destroying rare books to acquire training data. The practice involves acquiring physical books, scanning their contents, and discarding them—the cheapest and fastest method to capture high-quality long-form text for AI model training. Industry observers worry the practice is occurring on a larger scale than currently reported and that irreplaceable physical copies may be permanently lost. While companies view this as an efficient approach to sourcing valuable training material from rare or out-of-print books, conservationists and booksellers argue the strategy represents an unnecessary loss of cultural and historical artifacts.