Amazon Destroys Rare Books to Train AI Models

Investigative reporting revealed that Amazon purchases large volumes of printed books, including rare editions, scans them for its AI models, and then discards the physical copies after digitization.

How it works

Using an AirTag placed inside a shipment, 404 Media traced the package to Amazon’s Las Vegas warehouse labeled VGT3. Employees remove book covers to speed up scanning, which destroys the originals. The extracted data feeds Amazon’s Nova AI models offered to customers.

Legal reactions

An Amazon spokesperson said the books are bought through commercial channels to improve its services. Similar tactics were highlighted in a lawsuit against Anthropic’s “Project Panama,” where a court ruled the digitization qualified as fair use despite the original copies being damaged.

Broader context

The practice sparks debate over copyright and heritage preservation, especially when irreplaceable rare books are destroyed. Critics argue that turning such content into proprietary data for closed AI models deprives the public of access to the original works.

The issue raises questions about the compatibility of these methods with international law and cultural knowledge protection, as companies continue to expand data sets for AI training.