AI companies using destructive scanning for data training

Artificial intelligence companies are reportedly utilizing a process known as “destructive scanning” to acquire data for training large language models (LLMs). This practice involves the industrial-scale unbinding, shredding, or recycling of physical books after their pages have been scanned and converted into digital datasets.

As online digital knowledge becomes increasingly saturated with AI-generated content, tech firms are seeking authentic, non-digitized sources. Reports indicate that companies are placing massive, varied bulk orders at secondhand bookstores across the United Kingdom, Ireland, the United States, Canada, and mainland Europe. Investigations suggest that books—ranging from niche historical biographies to specific foreign-language editions—are being purchased and subsequently destroyed.

Recent scrutiny has focused on Amazon following an investigation by 404 Media. Using an Apple AirTag, investigators tracked a bulk order of approximately 1,000 books to an Amazon facility in Las Vegas, known as VGT3. Reports allege that employees at the facility remove the spines of books before scanning the pages, rendering the physical copies unusable.

While Amazon confirmed to City AM that it purchases books through commercial channels to “help develop and improve the products and services our customers use,” the company has not explicitly confirmed if the material is being used to train its Nova AI models. This follows similar reported practices by other developers, such as Anthropic.

Critics argue this treats books merely as containers for raw data, potentially undermining the physical medium as an artifact of human memory. From a legal standpoint, companies may use this method to claim “fair use” under U.S. law, as the original physical works are not being redistributed digitally.