Amazon destroys rare books to feed AI training models, investigation finds

Amazon is bulk-purchasing rare books and destroying them to digitize texts for AI training data. An investigation by 404 Media uncovered the operation, which utilizes a process known as "destructive scanning" to accelerate the ingestion of physical media into digital datasets.

To expose the operation, 404 Media planted an AirTag in a shipment of approximately 1,000 rare books. The tracking device followed the shipment from California through Milwaukee and Kenosha—where it stopped at a Trifinity warehouse—and Colorado before arriving in Las Vegas, Nevada. The books were delivered to the LAS8 warehouse, specifically within a unit designated as VGT3. At this facility, workers strip or cut the bindings from the books to facilitate faster scanning, effectively destroying the physical copies. The VGT3 team is identified by a logo featuring a T-Rex preparing to devour a book.

The race for high-quality data

This operation comes as AI developers face an escalating scramble for high-quality, human-written training data. As the industry exhausts easily accessible web-scraped content, demand has surged for digitized versions of physical texts. This has led to an increase in the bulk purchase of books through commercial channels to provide the complex linguistic patterns and deep knowledge found in rare volumes.

Critics have noted the irony of the practice: Amazon, a company Jeff Bezos founded as an online bookstore, is now permanently removing physical books from circulation to power its machine learning models. In a statement, an Amazon spokesperson confirmed the company's procurement strategy, stating, "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use."

Cultural and ethical costs

The practice raises urgent ethical concerns regarding the preservation of cultural heritage. By employing destructive scanning, Amazon prioritizes digitization speed over the survival of rare physical artifacts, potentially erasing unique copies of texts that cannot be replaced. Beyond the cultural loss, the process highlights the aggressive and often opaque methods AI companies employ to secure a competitive edge in data acquisition.

What remains unconfirmed

While the location and method of the scanning operation are now public, it remains unclear exactly which AI models are being trained with this specific data or the total volume of books destroyed to date. Industry observers are now watching to see if other AI labs are employing similar destructive tactics or if the backlash will force a shift toward non-destructive digitization standards.