AI companies are buying up pre-2022 printed books because the internet is too full of AI-generated text

ISBNdb, a book database company, is now offering bulk book acquisition services for AI companies, pitching pre-2022 printed books as ideal training data because they are "structurally guaranteed" to be free of AI-generated text. The company handles orders of 1,000 to 1 million books at a time and requires NDAs to keep AI companies' identities hidden — its own site notes that "'AI company destroys two million books' is not a headline that generates sympathy."

The pitch arrives as the web fills with AI-generated content, which can cause "model collapse" when used as training data. Booksellers on platforms like Alibris and Biblio report historic spikes in bulk purchases with "no rhyme or reason" — varied topics, disregard for price, and a focus exclusively on books with ISBNs. Internal Anthropic documents revealed in a copyright lawsuit detailed plans to scan millions of books and destroy them in the process; a federal judge ruled the copying was fair use specifically because the physical books were destroyed. One bookseller told 404 Media that rare, foreign-language, and low-circulation books are being swept up in these purchases, and if they're pulped during scanning, "they'll be even harder to obtain."