Tracking Device Uncovers Amazon's Rare Book Destruction for AI Training

An undercover investigation has provided the first concrete evidence that Amazon is systematically acquiring rare books in bulk, scanning them, and destroying them as part of its artificial intelligence training pipeline. The discovery, made possible by an Airtag secretly placed in a rare book shipment and tracked to an Amazon facility in Las Vegas, confirms what book dealers have long suspected: major tech firms are raiding the rare book market to fuel competitive AI development.

The Hidden Tracker Investigation#

404 Media partnered with a bookseller to plant a tracking device inside a rare book that was part of a large purchase order. The Airtag's signal led investigators directly to an Amazon warehouse in Las Vegas housing a specialized team responsible for extracting books from their bindings and digitizing pages. The operation, housed in a unit labeled VGT3, bore a particularly striking emblem on its door: a Tyrannosaurus rex in the act of consuming a book—a logo that seemed to mock growing public outcry over the practice.

Amazon's Vague Response#

When contacted for comment, Amazon refused to address the specifics of the investigation. Instead, the company issued a generic statement about purchasing books "through commercial channels to help develop and improve the products and services our customers use"—language that conspicuously avoids mentioning AI training. The non-answer underscores Amazon's apparent hesitation to openly acknowledge using rare books as training material as the practice faces mounting criticism from authors, publishers, and cultural advocates.

The Competitive Data Arms Race#

Amazon's aggressive sourcing of rare books reflects the high-stakes competition in frontier AI development. Companies racing to match rivals like Google, OpenAI, and Anthropic recognize that access to unique, hard-to-obtain training data offers a crucial edge. Rare books—by definition scarce and distinctive in content—represent untapped repositories of text that competitors may lack. This contrasts with public commitments made by firms such as Anthropic and xAI, which have explicitly stated they exclude rare and antique books from their training datasets.