Amazon is purchasing rare books, cutting off their spines, and scanning them to use as training data for its artificial intelligence models, according to a report by 404 Media published in August 2026.
404 Media confirmed the practice by placing a tracking device inside a rare book, which was subsequently purchased and transported to an Amazon facility in Las Vegas known as VGT3. The facility identifies itself with a symbol of a dinosaur holding a book in its claws. When contacted, Amazon told 404 Media that it “purchases books through commercial channels to improve the products and services customers use.”
The effort is driven by a growing scarcity of usable training data. AI companies need vast quantities of text to train large language models (LLMs), and much of what is available on the internet has already been consumed. Rare and out-of-print books — materials unlikely to be found online — represent a new source of text for that purpose.
These older texts carry an additional appeal: any book published before 2022 could not have been written by an LLM, making it free of a risk known as “model collapse.” That phenomenon can occur when an AI model’s output quality degrades after ingesting too much AI-generated text, making human-authored material particularly valuable for training purposes.
Amazon is not alone in its aggressive pursuit of training data. The source notes that Anthropic has faced scrutiny over the use of illegally pirated books for similar purposes. The destruction of rare books to feed AI systems may raise concerns among collectors, archivists, and scholars, though the full scope of Amazon’s book-acquisition program — including how many volumes have been processed — was not disclosed.
Source: TechCrunch