Amazon AI Training Facility Confirmed as Buyer and Destroyer of Rare Books

An Airtag hidden inside a rare book has traced a bulk order destined for systematic destruction directly to an Amazon artificial intelligence training facility in Las Vegas, confirming what booksellers had long suspected about the identity of the companies quietly consuming their inventory.

404 Media reported on Monday that it connected with a bookseller who agreed to plant an Airtag in a rare book included in a bulk order. The device was tracked to an Amazon warehouse, designated VGT3, where a team focused on removing book spines and scanning pages for AI training data operates, according to the outlet's findings.

A logo on the door of that warehouse depicted a Tyrannosaurus rex preparing to devour a book — the same image used as a team logo by the group working inside.

Amazon declined to address the findings directly. The company provided Ars Technica, which covered 404 Media's investigation, with the same statement it gave to 404 Media: "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use." The statement makes no mention of AI training specifically.

The disclosure arrives amid intensifying competition among technology companies to secure large volumes of original text for training frontier AI models. Rivals Anthropic and xAI have each publicly stated that they do not train on rare or antique books, a distinction that could offer Amazon a competitive advantage if its models benefit from the scanned material.

Workers at the VGT3 facility discussed their operations in online forums, where they indicated that earlier this year the facility had run low on books to scan — a shortage severe enough that some workers feared the warehouse might shut down. The supply had at one point completely run out, workers said, though the facility has since remained operational.

Those same online discussions revealed that workers were trained to scan barcodes or ISBNs before scanning book pages. That practice lends weight to a theory circulating among booksellers that AI firms are systematically working through the global ISBN registry to maximize the number of unique works present in their training sets.

The bookseller who planted the Airtag told 404 Media that the books being destroyed carry value that goes beyond their market price. "All sorts of things" determine a rare book's worth, the seller said, adding that "the AI companies don't care about. They just want the content as a bunch of words strung together."

Scottish bookseller Derek Walker told the BBC that the most troubling scenario would involve the destruction of a work that is "the only known surviving example of an edition." Walker said AI firms should distinguish between low-impact academic texts and irreplaceable antique works that have "survived this long."

Critics of destructive scanning note that non-destructive alternatives exist, making the physical elimination of books a choice rather than a necessity. Investor Michael Burry has reportedly labeled the practice "evil incarnate," though Amazon workers described the scanning gig in forums as a "nice" opportunity with flexible hours.

For booksellers, the revelation raises an ethical dilemma: bulk orders from AI firms may be financially attractive, but sellers now face the knowledge that carefully sourced collections may be fed into a scanning operation that treats their cultural and historical value as irrelevant.

As AI firms continue to guard their training data and obscure their identities as buyers, the tension between institutional book preservation and the data demands of frontier model development is unlikely to ease anytime soon.

Disclaimer