Amazon Is Shredding Rare Books to Feed Its AI Models

404 Media hid an AirTag inside a rare book, sold it into a bulk order of roughly 1,000 titles, and watched it travel across the country to LAS8 — an Amazon warehouse at 5801 Nicco Way in North Las Vegas. Inside sits a team called VGT3, whose logo is a T. rex clutching an open book. Workers describing the job on employee forums say that's the whole gig: take in pallets of books, scan the barcodes, slice off the bindings so pages feed through the scanners faster. The book does not survive.

Amazon's response was one sentence: it "purchases books through commercial channels to help develop and improve the products and services our customers use." It didn't address the destruction, the scale, the start date, or whether anyone screens out scarce editions before the blade. Outside coverage has linked the corpus to Amazon's Nova models; Amazon hasn't confirmed that.

The legal cover is real, and it's why this is happening in the open. A U.S. federal court has already blessed destructive scanning of legally bought physical books as fair use — the bright line courts drew was purchased vs. pirated, not intact vs. shredded. So with the open web scraped dry, buying paper and guillotining it is now the cleanest licensed-data play in the industry.

Why it matters: the cheapest legal path to fresh training data now runs through physical objects that don't come back.