Tracking rare books to be scanned and destroyed. The madness must stop. In a striking investigation on the hidden costs of the AI data Amazon has been revealed as a major buyer of rare and out-of-print books only to systematically dismantle them for machine learning. A bookseller who noticed an unusual surge in large bulk orders agreed to help track one shipment. In July 2026, as part of an order of roughly 1,000 volumes placed through the independent marketplace Biblio, an Apple AirTag was concealed inside a rare book. The tracker’s journey across the United States ended at an Amazon facility in Las Vegas, Nevada, known as LAS8 (5801 Nicco Way). Inside that warehouse operates a specialized team designated VGT3. Workers at the site have described their routine in online forums for Amazon employees: massive shipments of printed books arrive; some staff cut the bindings and spines off the volumes so pages can be fed more quickly through high-speed scanners; others handle receiving and barcode scanning. Once digitized, the physical books are discarded. The team’s internal logo: a Tyrannosaurus rex brandishing its teeth while clutching (or preparing to devour) a book has become a sick symbol of the process. “I work at VGT3 here in Vegas, and all we do is scan books,” one employee wrote. “Some are assigned to cut books, and others go to receive where they get books and scan the bar codes.” The facility had reportedly faced supply shortages earlier in the year severe enough that workers worried it might close; at one point the books ran out entirely. It remains operational. Booksellers had already observed what they described as a “historical spike in sales” of rare titles, leading many to suspect AI companies were quietly Hoovering up physical copies of works not readily available online. Printed books published before the generative-AI boom are particularly valuable: they contain high-quality human-written text free of the synthetic data that can contribute to “model collapse” when large language models train on their own outputs. Rare and out-of-print editions also expand the diversity of training corpora beyond what can be scraped from the open web.6 Amazon’s response has been carefully worded. A company spokesperson told investigators: “Amazon purchases books through commercial channels to help develop and improve the products and services our customers use.” The statement does not explicitly confirm that the scanned material is used for training the company’s Nova family of frontier models, though that is the clear implication drawn by observers given Amazon’s aggressive push into competitive large language models. The practice is not unique to Amazon. Similar destructive scanning operations have been documented or alleged at Anthropic and OpenAI. In a lawsuit involving Anthropic, a court found that when the company offered to purchase books, remove their spines, digitizing the pages, and destroying the originals it did not constitute copyright infringement. They had other methods. The irreversible destruction of physical cultural artifacts especially rare books raises ethical questions that go beyond strict legal fair-use analysis. Some have called the practice “evil incarnate.” Because it is. 404 Media article:
404media.co/we-tracked-a-shi…
 What began as a quiet suspicion among independent booksellers has now been confirmed by tracking device. The physical books that once defined Amazon’s early identity are, in at least one Las Vegas warehouse, being fed into the very systems that may one day make their printed form obsolete. AI companies should do the right thing and form a physical library for these books for the commons. Any court would agree, it is in the sprit of the law.