- Pre-2022 print books are a clean source of AI training data, free from AI-generated text and poisoning.
- ISBNdb offers bulk book acquisition services for AI companies, with strict NDAs to keep client identity secret.
- Bulk purchases of books by AI companies have spiked since 2024, causing unusual sales patterns recognized by booksellers.
- Destructive scanning of books for AI training is faster and cheaper, and has been ruled as fair use in a lawsuit.
- Books sold for training data are from the secondary market, giving no compensation to original creators.