We Tracked a Shipment of Rare Books. It Ended at an Amazon AI Training Facility

In July, 404 Media reported on booksellers seeing a historical spike in sales over the past year. They suspected AI companies were acquiring any books they could find to train their models, because buyers ordered massive quantities, chose seemingly random titles, and were not price sensitive — unlike libraries or universities. However, sellers couldn't confirm who the buyers were because the marketplaces that process orders keep buyer identities anonymous.

The investigation placed a tracking device inside a rare book shipment suspected to be destined for an AI company and followed it across the country. It ended at an Amazon warehouse in Las Vegas, Nevada, called VGT3. According to Amazon employees who work there, the facility receives massive shipments of printed books and scans them for AI training data. To speed up scanning, workers cut the bindings off the books, destroying the printed book in the process. The VGT3 team's logo is a dinosaur with a book in its hands.

Amazon confirmed the operation in a statement: "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use." This is the first known confirmation that Amazon is scanning and destroying books for AI training. Printed books are valuable training data because much of their text is not freely available online, and books published before 2022 are guaranteed to be free of AI-generated text, which otherwise can degrade models through a process called "model collapse."

The report confirms that at least one major AI company is physically destroying books to obtain training data — a practice that has drawn public criticism. It also underscores how far AI companies are willing to go to secure fresh, non-AI-generated source material as they exhaust other text sources.