Amazon Is Buying and Shredding Rare Books to Train Its AI Models

A rare bookseller who suspected an unusually large order was headed somewhere other than a reader's shelf slipped an Apple AirTag between the pages of one volume before shipping it out. That tracker led straight to a fenced Amazon facility outside Las Vegas, where workers say their entire job is cutting the spines off incoming books so the pages can be run through a scanner faster, destroying each copy in the process, according to an investigation by 404 Media.

What the investigation found

The order that triggered the tracking came through Biblio, a marketplace used by secondhand and antiquarian booksellers, and reportedly ran to around a thousand books. The shipment ended up at a unit Amazon workers refer to internally as VGT3, marked with a logo of a dinosaur devouring a book. One employee, posting on a forum used by Amazon staff, put it bluntly: "I work at VGT3 here in Vegas, and all we do is scan books." Reporting from TechCrunch and Futurism corroborates the same picture: bulk deliveries of printed books arrive, get sliced apart for faster scanning, and the physical originals do not survive.

Asked about the facility, Amazon did not deny that the books feed its AI efforts. The company told 404 Media only that it "purchases books through commercial channels to help develop and improve the products and services our customers use." One bookseller quoted in the piece put the buyer's motive more plainly: AI companies "just want the content as a bunch of words strung together" and have no interest in a book's condition, edition, or history, the details that usually matter most to a collector.

Why rare and out-of-print books specifically

Books that were never digitized, or exist only in small print runs, are attractive training material precisely because they have not already been scraped from the open internet a hundred times over. As frontier AI models run short of fresh, high-quality text, older and rarer books become a distinct kind of resource, one that a company can simply buy on the open market rather than negotiate a license for. Coverage from NPR frames this as a wider pattern, not a one-off: several AI developers now appear to be quietly competing for the same pool of physical, previously undigitized books.

This follows closely on the heels of reports that secondhand bookshops across the UK and Ireland were seeing unexplained bulk orders, which booksellers suspected but could not confirm were tied to AI training. The 404 Media investigation is the first time a specific buyer, Amazon, has been tied directly to a physical scanning operation, with a documented trail from bookshop shelf to warehouse floor.

What this means for you as an author

If your book is out of print, sitting in a small press run, or only available secondhand, this is a reason it might end up training a language model with no notice to you and no payment. A 2025 US court ruling already found that training an AI model on lawfully purchased books can qualify as fair use, so a purchase like this sits in a legal gray area that currently favors the buyer, not the author whose words are inside the book. There is little you can do to stop a specific copy of your book from being bought and scanned this way.

What you can do is be deliberate about how you talk about your own use of AI, since author transparency is becoming as much a part of the conversation as company transparency. Our guide on AI writing ethics for authors is a useful starting point if you want a clear answer ready the next time a reader or publisher asks how a book of yours came together.

If you are working on a new book and want a tool that keeps your voice and your process on your side of the table, PageWriter Studio builds from your own style profile and keeps a clear record of what you drafted versus what AI assisted with, so the authorship of your work is never in question. You can start a free trial to see how it works.