AINEWS
404 Media followed a single order from used book marketplace Biblio to a Las Vegas facility, where workers say books are sliced, scanned, and discarded for training data.
An Apple AirTag slipped into a single book let 404 Media physically trace a Biblio marketplace order of about 1,000 used books from an independent bookseller to a specific corner of an Amazon warehouse in Las Vegas. The tracking ended at a corner workers say is used to destructively scan large volumes of books.
The package landed at the VGT3 corner of LAS8, an Amazon warehouse in the northeast of Las Vegas, a building marked at the entrance by a dinosaur reading a book. The dinosaur is branding. The address is evidence. VGT3 is where the tracking stopped, and where, according to discussions between Amazon workers on an internal forum as cited by 404 Media, large volumes of books are run through a destructive-scanning process: high-speed machines slice the spine off each book, scan the loose pages, and pulp or discard the paper. The mechanics of the cut-and-digitize pipeline are described in Phandroid's July 2026 re-report.
The triggering order was a single Biblio transaction in July 2026 from a price-insensitive, anonymous buyer, a pattern independent booksellers had already flagged in dealer communities as the early signal that a lot was headed into the AI training supply chain. The price-insensitivity is the tell: a collector or a library replacement order haggles. A training-data buyer pays retail and asks for the lot.
This is the first on-the-record physical trace of a specific shipment to a named Amazon facility. Prior rounds had inferred the scale from secondhand accounts and supply-chain patterns, in coverage by the Decoder and Futurism, and in the Anthropic book-scanning story 404 Media cites as the prior round. The 404 Media AirTag method converts that inference into a pin on a building.
The mechanism delta is what changes the category. A sourcing investigation built on documents and secondhand accounts can be denied or reframed. A package with a working GPS tracker that ends up at a specific loading dock at a named facility cannot. The method is the part that travels: a thirty-dollar-class device, a willing bookseller, and a known destination.
Both the bookseller and the worker-forum accounts place the VGT3 corner at the center of a destructive-scanning operation. The AI-training purpose is the source community's inference, not an Amazon confirmation. The corner is the fact. The training-data label is the reading.
The unresolved questions are upstream and downstream. Upstream: how a Biblio order from an anonymous buyer translates into Amazon's procurement pipeline, and whether the booksellers on the other end understood what their inventory was feeding. Downstream: the open question of consent and copyright for books that are sliced, digitized, and discarded, before any model is trained on them. The earlier reporting rounds had framed those questions as live. The 404 Media investigation adds the physical receipt, and an address where the next round of source work can start.
Reader reaction to the report landed on a Hacker News thread dominated by the question of what counts as a "rare" book in a typical bulk bookseller order. The order was a mixed used-book lot, not a curated rare-books collection, and the framing in some of the secondary coverage tends to overstate the title.