An approximately 1,000-book shipment ordered through Biblio and tracked with an Apple AirTag reportedly ended at Amazon’s LAS8 facility near Las Vegas in July 2026. Accounts from employees describe an internal operation called VGT3, where bindings are cut away, loose pages are scanned, and the physical books are discarded. The reporting strongly connects the process to the race for AI training data, but it does not identify a specific Amazon model, dataset, or downstream path for the scanned text.
The shipment tracked to LAS8
The AirTag established a route and a destination: a book from the large shipment reached Amazon’s LAS8 facility in the Las Vegas area. That is important evidence, but it is not a complete chain of custody for the tagged book. The device did not, by itself, prove what happened to that particular copy after it arrived or show that its pages entered a named AI system.
The broader story also includes a separate reported order of 5,000 books described by bookseller Tomás Kenny. That order is not the same as the approximately 1,000-book shipment tracked with the AirTag, so the two figures should not be combined into a single total.
Inside the reported VGT3 workflow
Employees reportedly described VGT3 as a high-volume book-processing operation. Each book is identified by its ISBN or barcode. Workers then cut through the spine or binding so the pages can be separated and fed through industrial scanners. Once the pages are loose—and, according to one worker account, mixed together—the physical book cannot be reassembled in its original form.
The practical consequence is straightforward: the process creates a digital scan while sacrificing the physical copy. An anonymous VGT3 employee summarized the operation bluntly: “All we do is scan books.”
The selected documentary footage shows the tracked route, the LAS8 facility, the VGT3 emblem, and a close-up demonstration of a book spine being cut. It documents the investigation’s central visual clues, not an exhaustive audit of Amazon’s operations or proof that a particular model received the scanned pages.
What the VGT3 emblem tells us
The VGT3 emblem depicts a dinosaur tearing into an open book. It is a distinctive visual identifier for the reported operation, although a logo cannot establish the full purpose of a facility or prove how its output is used.
Amazon’s response stops short of naming AI
Amazon’s supplied public statement says it “purchases books through commercial channels to help develop and improve the products and services our customers use.” That confirms the company’s stated reason for buying books, but it does not confirm the reported VGT3 workflow, explicitly mention AI training, or identify a model, dataset, or data pipeline.
So, does the evidence show that Amazon destroys books for AI? It supports a reported process in which books connected to Amazon’s LAS8/VGT3 operation are cut apart and scanned, alongside reporting that frames the activity as part of AI-data collection. The precise connection between those scans and a specific Amazon AI system remains unconfirmed.
The interview below provides additional context about the AirTag investigation, the facility, the destructive process, and Amazon’s limited statement. It does not identify a model or establish that every Amazon book purchase follows the same path.
“Rare” does not automatically mean irreplaceable
The books involved have been described variously as rare, old, out of print, or limited-circulation. Those labels are not interchangeable. A hard-to-find book may still have many surviving copies, while a genuinely unique edition would raise a much more serious preservation concern.
The supplied evidence names no specific title destroyed at VGT3 and does not establish that an irreplaceable surviving edition was processed there. That distinction matters: the physical destruction described in the employee accounts is significant, but it should not be inflated into proof that a cultural last copy was lost.
It also explains why headlines about “rare books” can mislead. Without a title-by-title inventory, readers cannot determine whether the shipment contained collectible antiques, obscure academic books, ordinary out-of-print editions, or a mixture of all three.
Google Books shows another technical path
Destructive scanning is not the only way to digitize books. The Google Books comparison describes specialized scanning that allowed books to be returned without cutting their bindings. It demonstrates that non-destructive book digitization is technically possible.
That comparison does not establish which method is faster, cheaper, or capable of handling more books at Amazon’s scale. It does, however, remove one tempting shortcut from the debate: cutting a book apart is not the only technically available approach.
| Approach | Physical handling | AI or digitization relationship |
| Reported Amazon VGT3 workflow | Bindings are reportedly cut away; loose pages are scanned and physical copies discarded. | Reporting connects the operation to AI-training data, but no specific Amazon model or dataset is identified. |
| Anthropic’s Project Panama | Court-related reporting describes destructive scanning of legally acquired printed books; a separate dispute involved pirated digital books. | The project was connected to AI-model training in the related litigation. It is a separate case, not proof of Amazon’s legal position or data pipeline. |
| Google Books comparison | Specialized scanners were used without cutting bindings, allowing books to be returned. | A large-scale digitization example showing that physical destruction is not technically necessary in every scanning workflow. |
What the Anthropic comparison changes—and what it does not
Anthropic’s Project Panama provides context for the wider fight over training data, copyright, and preservation. The supplied court materials distinguish legally acquired books from pirated copies, and court-related reporting describes destructive scanning in that separate project.
But that does not turn Anthropic’s case into a ruling about Amazon. Nor does it establish a blanket rule that U.S. copyright law requires every legally purchased original to be destroyed after scanning. The legal materials supplied here do not support that categorical claim.
The cultural question remains more concrete than the legal slogan: when a physical book is cut apart for digitization, what value is lost, and who decides that the loss is acceptable? In the Amazon case, the reported workflow is clear enough to explain the concern. The identity of the final AI recipient—and the value of each individual book—still requires careful separation from assumption.
Bottom line: the AirTag investigation reportedly tied an approximately 1,000-book shipment to Amazon’s LAS8 facility, while employee accounts described VGT3 cutting bindings and scanning loose pages. Amazon confirmed commercial book purchases, not a named AI model or dataset. The strongest conclusion is therefore narrower—and more useful—than the viral headline: a destructive book-scanning operation has been reported, but its exact AI pipeline and the identity of any destroyed titles remain unspecified.
