Amazon destroys rare books for AI training, 404 Media tracking device exposes operation

An investigative report from 404 Media revealed that Amazon is purchasing rare books, removing their spines, and scanning them to train large language models. The outlet hid a tracking device inside a rare book, which eventually arrived at Amazon's VGT3 facility in Las Vegas, a building marked with a logo of a dinosaur holding a book in its claws. When confronted, Amazon said it “purchases books through commercial channels to improve the products and services customers use,” effectively confirming the books are being used for AI training data.

Rare and out-of-print books are especially valuable because they contain text unseen on the internet and, crucially, were published before 2022, meaning none of it was written by an AI. Feeding LLMs AI-generated output risks “model collapse,” where the model's output quality degrades after training on too much synthetic text. This makes pre-2022 books a premium resource for companies racing to build more capable models. But the practice destroys physical copies of often irreplaceable works, raising serious concerns about cultural preservation and the ethics of consuming scarce archival materials for commercial AI projects.