AI burns books? Tech giants are buying, shredding books to train smarter chatbots

AI companies are reportedly purchasing and destroying millions of rare books to train new AI models and prevent "model collapse" and "AI slop." These tech giants pay contractors to acquire books, which are then high-speed scanned, with their spines cut and originals shredded. ISBNdb, a company claiming the world's largest book database, stated that pre-2022 books are ideal for AI training as they are free of AI-generated text. This practice has drawn criticism, with former US House Rep. Brad Carson noting, "The law now rewards destruction and penalizes preservation."

• Anthropic, backed by Amazon and Alphabet, faced a multi-million dollar class-action lawsuit from authors whose pirated books were used to train its chatbot, Claude.
• A judge ruled that Anthropic's destruction of the physical books made their use of digital copies for training a "fair use" transformation, not a copyright violation.
• Internal documents reveal companies' desire to keep this practice secret, with one stating, "We don't want it to be known that we are working on this."
• ISBNdb has since denied purchasing, scanning, or selling books for AI training, stating a marketing page was merely a "test of market interest."