Amazon linked to destructive rare-book scanning

A hidden tracking tag inside a rare book led a bookseller’s shipment to an Amazon facility in Las Vegas where workers tear books from their spines and scan the pages, according to Ars Technica citing reporting by 404 Media. Booksellers have suspected for about a year that AI firms were placing bulk orders for hard-to-find titles and destroying them after scanning, and the Airtag trail offered a concrete destination.

Ars Technica reports that the Las Vegas site is tied to Amazon’s AI training effort, with a small team doing high-volume scanning work. One detail from the report captures the industrial tone: a door marked with a Tyrannosaurus rex logo poised to devour a book. Amazon declined to comment on the AI-training claim specifically, but said it buys books through commercial channels to develop and improve products and services.

The episode sits inside a wider scramble for proprietary training data. Companies building frontier models increasingly treat unique text as an input worth paying for, while keeping sourcing practices opaque. Ars Technica notes that other firms, including Anthropic and xAI, have said publicly they do not train on rare or antique books. That leaves Amazon’s approach looking less like preservation-minded digitisation and more like supply-chain optimisation: buy what is available, extract the text, discard the physical object.

The reporting also sketches a method behind the buying. 404 Media suggested the purchases may be driven by ISBN lists, a way to systematically seek out unique works rather than famous titles. Workers were trained to scan barcodes or ISBNs before scanning pages, according to the article, aligning the workflow with cataloguing rather than curation. In online forums, workers reportedly worried earlier this year that the facility might shut down if it ran out of books to scan; at one point, the supply dried up completely. Ars Technica says the facility remains operational.

For booksellers, the market signal is blunt. Bulk buyers appear to skip the usual judgments about condition, edition, or cultural value and instead pay for uniqueness at scale, pushing sellers into an ethical trade: accept the money and watch the inventory vanish, or refuse and lose the sale to someone else. The practice also shows how “data acquisition” can look when it is cheaper to destroy a book than to negotiate rights, handle careful scanning, or store the object afterward.

The Airtag’s final ping was not a library or an archive. It was a warehouse built to turn bound pages into model input, one spine at a time.