Amazon has been caught buying rare books in bulk and destroying them to extract their contents for AI training, after investigators planted a hidden tracking device inside one of the shipments.
The operation was uncovered when reporters at 404 Media persuaded a bookseller to conceal an Apple AirTag within a large batch of books that Amazon had ordered anonymously. The tag led investigators to a facility known as VGT3, an Amazon AI training warehouse in Las Vegas, where books have their spines removed so the pages can be scanned before the physical copies are discarded. The warehouse door carries a sign featuring a dinosaur logo that appears to show the creature poised to consume a book.
Drawing on the 404 Media investigation, found that workers at the facility are required to scan barcodes or ISBNs before processing each title, pointing to a systematic effort to work through as broad a range of printed material as possible.
The appeal of older physical books comes down to provenance. Titles published before the AI era carry a built-in guarantee of human authorship, which means there is no risk that training data could be contaminated with AI-generated text, a known problem that can degrade model quality.
Amazon is not the only company pursuing this approach. Anthropic, the maker of the Claude AI assistant, has reportedly spent tens of millions of dollars acquiring millions of books to satisfy its own demand for human-authored material, according to earlier Washington Post reporting.
Concerns had reportedly circulated among Amazon staff that the facility could face closure if the supply of printed books dried up. Ars Technica has since reported that the warehouse remains operational.
In response to the findings, Amazon stated that it purchases books through commercial channels to help develop and improve the products and services its customers use. The company made no comment on the destruction of the volumes.
The disclosure adds to mounting scrutiny of how AI developers source training data, with rare and out-of-print books becoming a particularly sought-after commodity precisely because their age guarantees what AI-generated text cannot: that a human wrote every word.
