The company that began as an online bookseller is now facing a stark accusation: that irreplaceable printed works are being sacrificed to feed its AI ambitions.

The trail began with a shipment of rare books fitted with a tracker by 404 Media. It led to VGT3, an Amazon facility in Las Vegas, where the warehouse identifies itself with an image of a dinosaur clutching a book. The investigation reported that the books were not being preserved or resold, but processed for their text.¹

Workers at the site told 404 Media that they receive large consignments of printed volumes, then “cut the bindings off in order to scan the books more quickly.” The consequence, they said, is blunt: “The printed book is destroyed in the process.”²

The reported practice reflects the escalating scramble for clean, human-written material to train large language models. Internet text has been heavily harvested, while older and out-of-print books offer material that is both scarce online and safely predates the flood of AI-generated writing. That distinction matters because models trained too heavily on synthetic output can suffer “model collapse,” degrading the quality of what they produce.¹

Amazon did not dispute that it acquires books, but offered a far less dramatic framing. The company said it “purchases books through commercial channels to improve the products and services customers use.”¹ The gap between those accounts is the heart of the controversy: Amazon describes an ordinary commercial input; critics see a one-way conversion of rare cultural objects into machine-readable data.

What the tracker establishes is not merely an AI industry hunger for text, but where that hunger may now be reaching: shelves of books too obscure, old or unavailable to have survived online.