Amazon scans and destroys rare books for AI training data

Companies developing large language models (LLMs) require vast amounts of text data. Physical books, especially older or out-of-print ones, are seen as valuable sources of clean, human-written text, helping to avoid "model collapse" which can occur when AI models ingest too much AI-generated content. Amazon stated to 404 Media that it "purchases books through commercial channels to improve the products and services customers use."

Similar practices have been reported for other AI companies, with Anthropic facing criticism for its "Project Panama" initiative, which involved buying millions of books, removing their bindings, scanning them, and disposing of the originals. While digitizing books can preserve knowledge, the destructive nature of this scanning process has raised concerns about the permanent loss of unique physical artifacts, such as first editions, handwritten notes, and original bindings, which hold historical and cultural significance beyond their text.