Amazon is purchasing and destroying rare books by cutting off their spines and scanning them, using the text to train its large language models (LLMs), according to an investigation by 404 Media.

The outlet placed a tracking device in a rare book that eventually arrived at Amazon facility VGT3 in Las Vegas, which displays a dinosaur holding a book.

Amazon told 404 Media that it “purchases books through commercial channels to improve the products and services customers use.

” The article explains that LLM developers need enormous volumes of text beyond what is available on the internet (and notes that Anthropic previously illegally pirated books).

Rare books that are out of print or not digitized offer a fresh source of training data, crucially containing only human-written text from before 2022.

Training on AI-generated content risks “model collapse,” where output quality degrades, making these pre-LLM texts especially valuable.

The piece does not specify how many books Amazon has acquired or how long the practice has been ongoing.