AirTag confirms Amazon is destroying books to train AI

Booksellers have spent the past year fielding strange, oversized orders for used books with no apparent connection between titles and no obvious buyer behind them.

Many suspected the orders were coming from AI companies looking to scan the books for training data. An investigation from 404 Media, which planted a tracking device inside one such shipment, has now traced that suspicion directly to an Amazon warehouse in Las Vegas.

Working with an anonymous seller on the online marketplace Biblio, 404 Media embedded an Apple AirTag into one book out of a batch order of about 1,000 books that the vendor thought would be used to train AIs.

The AirTag moved across California, through Milwaukee; stayed for two weeks around Kenosha, Wisconsin; and travelled via Grand Junction, Colorado, before reaching LAS8, an Amazon warehouse located in the northeast part of Las Vegas.

VGT3 was the name for the specific area of the LAS8 warehouse where the AirTag’s signal originated from. The area was marked with the company logo featuring a dinosaur opening its mouth above an open book.

The employees confirmed to 404 Media that huge deliveries of physical books arrive there, and the book spines are cut off to scan them faster.

Amazon told 404 Media only that it "purchases books through commercial channels to help develop and improve the products and services our customers use", a statement that does not mention AI directly.

There are some unique advantages in using used and rare books for training AI models in that any book which predates the common usage of generative AI would definitely have been written by a human being.

Publishing companies usually do not allow their authors to include any generated content, but there are no such restrictions on self-published books.

It has been speculated by booksellers that buyers are going through all the ISBN numbers to get copies of every book possible.