From book to dataset
A worker's account of the process at Amazon's VGT3 facility in Las Vegas is brief and irreversible: machines cut the spines from books, the loose pages pass through about 20 to 25 high-speed scanners, and the paper then goes into large boxes. The book can no longer be reassembled while its text becomes a portable digital copy. The destruction of the physical object is visible; the copy's next journey is not.[1]
The same line takes in sealed new books, volumes from library liquidations, works in German, Russian and Japanese, and University of London documents. That variety does not place every source under the same terms of use. Bulk acquisition is as plausible as deliberate selection of particular titles; either way, the warehouse process says nothing about the rights basis for each book.[1]
The missing link is the digital copy's recipient
In earlier reporting, 404 Media followed a tracking device placed in a shipment of rare books to VGT3. The new interview describes how scanning and destruction happen inside the facility. The outlet still could not identify which AI company receives the scanned data. The physical chain therefore has a known destination while the first recipient in the digital chain remains unknown.[1]
Amazon did not respond for the report, and that silence cannot carry a finding of misuse. Yet an unnamed recipient also cuts off the route by which an author, publisher or library could learn which model used a work, how long the copy is retained, or whether it moves to another organisation. The report provides no document answering those questions. The uncertainty sits at the digital copy's exit from the warehouse, rather than the book's arrival there.[1]
Disclosure should travel with the page
A useful disclosure names the entity receiving the scans, explains how books are selected, and states the rights basis used for each kind of source. The same chain includes the retention period, uses beyond model training, transfers to other entities, and a deletion route. A list of scanned titles produces an inventory. Connecting each title to the controller of its digital copy gives an affected person a path to challenge the use and seek a remedy.[1]
The signal to watch is narrow: a document connecting the VGT3 scans to a named data controller, a stated purpose, and a published retention period. The visible waste in Las Vegas is cut paper. The durable data exhaust can travel much farther than the book that can no longer be reassembled. Without a named recipient, the chain's most consequential handoff is also its least visible link.[1]