A 404 Media investigation using a hidden Apple AirTag has revealed that Amazon is purchasing bulk shipments of rare books, cutting off their spines, and scanning them to train its AI models. The tracker led investigators to Amazon's VGT3 facility in Las Vegas, contradicting Amazon's prior denials of destructive scanning. AI companies highly prize pre-2022 printed texts because they are free from AI-generated content and adversarial data-poisoning. Amazon's operation mirrors Anthropic's controversial Project Panama, which a federal judge ruled was protected under fair use.
AirTag tracking investigation results
- ▪The 404 Media investigation was initiated in July 2026 when a bookseller on the independent marketplace Biblio agreed to place an AirTag inside a rare book within a bulk order.
- ▪An investigation by 404 Media tracked a bulk shipment of approximately 1,000 books containing a hidden Apple AirTag to an Amazon facility in Las Vegas, Nevada.
- ▪The AirTag traveled through Milwaukee and Kenosha, Wisconsin, and Grand Junction, Colorado, before transmitting its final signal from the north end of Amazon's LAS8 warehouse complex in Las Vegas.
VGT3 book scanning facility
- ▪VGT3 employees are trained to scan each book's ISBN barcode before scanning its pages, allowing Amazon to catalog the specific titles it has processed.
- ▪Amazon workers reported in online forums that the VGT3 facility ran completely out of books to scan earlier in 2026, sparking concerns that the warehouse might shut down.
- ▪The VGT3 facility features an internal logo depicting a Tyrannosaurus rex holding a book in its claws.
- ▪Amazon operates a specialized unit coded as VGT3 inside its LAS8 warehouse complex in Las Vegas, where employees cut the spines off books to scan pages into high-speed scanners.
Amazon prior denial contradicted
- ▪Amazon declined to comment on the specific findings of the 404 Media investigation, stating only that it purchases books through commercial channels to improve customer products and services.
- ▪The AirTag tracking results directly contradicted Amazon's prior denials that the company was engaging in destructive book scanning.
Pre-2022 book training value
- ▪Pre-2022 printed books are free from adversarial data-poisoning corruptions, such as those designed by the University of Chicago's SAND Lab, which can degrade AI model training.
- ▪Epoch AI researchers project that the available supply of public human-generated text for training large language models could be exhausted between 2026 and 2032.
- ▪Books published before 2022 are highly valued for AI training because they are guaranteed to be free of AI-generated content, protecting models from model collapse.
Destructive scanning for AI
- ▪Anthropic previously operated a book-scanning initiative called Project Panama, led by former Google Books executive Tom Turvey, to destructively scan books for AI training.
- ▪Anthropic finalized a $1.5 billion settlement in July 2026 to resolve class-action copyright claims regarding its unauthorized use of 7 million pirated books.
- ▪In the lawsuit Bartz v. Anthropic, U.S. District Judge William Alsup ruled in June 2025 that converting a lawfully purchased physical book into a digital file for AI training is transformative fair use.
Story comments
Loading comments…