Amazon is buying printed books in bulk, cutting the bindings off, scanning the pages for AI training data and discarding the paper, according to a 404 Media investigation published on 17 August 2026. Reporter Emanuel Maiberg confirmed the operation by hiding an Apple AirTag inside one book in a 1,000-title order and tracking it across the United States.
The tracker's final stop was LAS8, an Amazon warehouse in north-east Las Vegas, Nevada. The book-scanning unit inside LAS8 is called VGT3, and the logo at its entrance is a Tyrannosaurus rex holding an open book.
What the AirTag Investigation Found
The 1,000-book order was placed through Biblio, a rare and used book marketplace that keeps buyer identities anonymous. An anonymous bookseller working with 404 Media planted the AirTag, which moved through several states before arriving in Las Vegas.
Amazon employees describing VGT3 on worker forums said their entire job is receiving pallets of books, scanning barcodes and cutting bindings so the pages feed through scanners faster. Amazon told 404 Media it "purchases books through commercial channels" to improve its products, and named no model. Secondary coverage of the investigation has tied the data to Amazon's Nova model family; Amazon has not confirmed that.
Why AI Companies Want Printed Books
Printed text published before 2022 is the scarcest training resource in AI right now. Much of it was never digitised, so it escaped the web scrapes every large model has already consumed — and nothing printed before 2022 can contain AI-generated text. That second point matters because training a model on other models' output degrades it, a failure researchers call model collapse.
Booksellers told 404 Media the buyers scan every ISBN (the serial number unique to each edition), which supports their theory that AI firms are working through printed books by serial number.
How Amazon Compares With Anthropic's Project Panama
Amazon is not the first company to do this. Court exhibits in Bartz v Anthropic PBC described an internal effort to "destructively scan all the books in the world", codenamed Project Panama. Judge William Alsup ruled that scanning lawfully purchased books to train a model was transformative fair use, partly because the printed original was destroyed rather than resold.
| Detail | Amazon (VGT3) | Anthropic (Project Panama) |
|---|---|---|
| Revealed by | 404 Media AirTag investigation, 17 Aug 2026 | Court exhibits, Bartz v Anthropic PBC |
| Site | LAS8 warehouse, Las Vegas, Nevada | Third-party digitisation vendors |
| Method | Bindings cut, pages scanned, originals discarded | Spines cut, pages trimmed, originals discarded |
| Legal status | No court ruling as of 18 August 2026 | Ruled transformative fair use by Judge Alsup |
What This Means in India
No Indian court has ruled on destructive scanning. On 24 July 2026, Justice Amit Bansal of the Delhi High Court refused ANI Media an interim injunction against OpenAI, holding on a prima facie basis that storing news content to train a large language model falls under the fair dealing exception in Section 52(1)(a) of the Copyright Act, 1957. India's exceptions are a closed statutory list rather than the flexible US fair use test, and the physical destruction of a copy was never at issue there. The main ANI suit continues.
Quick Answers
Which Amazon facility scans books for AI training?
LAS8, an Amazon warehouse in north-east Las Vegas, Nevada. The scanning unit inside it is called VGT3.
Does Amazon destroy the books it scans?
Yes. Workers at VGT3 told 404 Media the bindings are cut off to speed up scanning, which destroys the printed copy.
Is destructive book scanning legal in India?
No Indian court has ruled on it as of 18 August 2026. The Delhi High Court's July 2026 interim order covered AI training on text, not the destruction of physical books.
The bookseller who worked with 404 Media put the objection plainly: "They just want the content as a bunch of words strung together."