An Apple AirTag was tucked inside a used book, shipped out as part of an order of roughly 1,000 books from a bookseller. Its signal eventually stopped inside an Amazon warehouse in northeast Las Vegas, facility code LAS8, in a specific zone called VGT3. 404 Media turned the tracking into an investigation.
A sign at VGT3's entrance shows a dinosaur holding a book. Workers there say the site does one thing: take in large volumes of printed books, cut off the spines, scan the pages, and discard the paper originals afterward. Discussions among Amazon employees on online forums back this up — VGT3 runs destructive scanning, at scale.
How a thousand books got bought
The books moved through Biblio, an online marketplace specializing in used and rare books. In July, a seller received an unusually large order — around 1,000 books — from a buyer whose identity gave nothing away. A reporter contacted the seller and asked them to slip a tracker into one of the books; the rest was left to shipping.
Amazon's only response was one line: the company "purchases books through commercial channels to improve the products and services customers use." The statement doesn't deny any part of the story — it just files the whole operation under "improving products."
Why paper books suddenly got valuable
A model company buying paper books wouldn't have happened a few years ago. The public web's supply of clean, long, well-structured text has largely been scraped through in earlier training rounds. The bigger problem is the timeline: internet text from 2022 onward is increasingly mixed with content the models themselves generated, and training on that degraded material makes outputs worse with each round — the industry calls this model collapse. That's why many training datasets deliberately cut off around 2022, to make sure what goes in was written by humans.
Paper books sit right in that gap. They carry a clear publication date, editor-proofed text density, and are certainly not machine-generated. Out-of-print and rare titles add another edge: their content has never existed online in digital form, so whoever scans it first owns an exclusive corpus.
That also explains the insistence on buying rather than pirating. The same text pulled from a pirated ebook library carries an entirely different level of legal risk.
A judge drew the line
In June 2025, Judge William Alsup of the U.S. District Court for the Northern District of California split the question down the middle in Bartz v. Anthropic. Paper books Anthropic paid for, then cut apart, scanned, and discarded, counted as fair use — the court's reasoning was that this simply converted one paper copy into one more convenient digital copy, without multiplying copies or distributing them externally. Anthropic had run this internally under a codename, Project Panama, building the library since early 2024.
The same ruling went the opposite way on the other half of the case: ebooks obtained through piracy are not protected by fair use. Anthropic later paid $1.5 billion to settle that part — the largest copyright settlement figure in U.S. history.
Put together, the two halves of the ruling amount to an operating manual for the entire industry: if you want the text inside books, buy the paper copies on the secondary market. What Amazon is doing in Las Vegas fits squarely inside that manual.
A company that started out selling books
There's an inescapable comparison here. Amazon got its start selling books in 1995, and Kindle has since given it one of the largest ebook licensing networks in the world. In theory, it has a more straightforward route to text: negotiate licensing directly with publishers. Choosing instead to buy up used books in bulk and destroy the paper afterward suggests that licensing route isn't competitive on cost or speed.
Publishers have no seat at this table. Once a book has been sold once, it belongs to someone else — authors and publishers get no second payment, and no way of knowing whose training set their book ends up in. Google Books' large-scale library scanning years ago drew its own controversy, but at least those books went back on the shelves afterward. This batch won't be going back.
Sources: 404 Media investigation, TechCrunch, CocoLoop, public court filings from the U.S. District Court for the Northern District of California; order size, facility code, and Amazon's statement verified against reporting; settlement figure verified against court records.