Back to News
NewsJobsAiFederal Reserve

Investigation Reveals Amazon's Destructive Book Scanning for AI

An investigation reveals Amazon's VGT3 facility destroys books for AI training, raising ethical concerns and sparking legal challenges.

Mercation Admin
August 18, 2026 1 min read
Investigation Reveals Amazon's Destructive Book Scanning for AI

Amazon's Book Scanning Practices Exposed

An investigation has revealed that Amazon's VGT3 facility in Las Vegas is at the center of a controversial practice involving the destruction of books for AI training. This inquiry began when an Apple AirTag was hidden inside a significant shipment of rare books, allowing for tracking to the Amazon facility.

At VGT3, employees are tasked with dismantling the spines of these books to expedite scanning processes, resulting in the irreversible destruction of the physical texts. In response to these findings, Amazon admitted to purchasing books through commercial channels but noted that this operation had not been disclosed to the public previously.

AI Training and the Demand for Printed Material

The investigation highlights the growing demand for pre-2022 printed books, which contain text not widely available online and are essential for training AI models. Such materials are crucial as training AI solely on its outputs can lead to "model collapse," compromising the quality of the models. Consequently, companies are increasingly sourcing these physical books, which are often scanned and discarded after use.

Amazon's practices are not isolated; a larger industry has emerged focused on providing AI companies with physical books for similar purposes. This trend has drawn comparisons to the dystopian themes of Ray Bradbury's "Fahrenheit 451," where books are destroyed to serve machine learning needs. Notably, Anthropic's "Project Panama" is already digitizing millions of books through this destructive method.

Legal Implications and Industry Reactions

The quest for high-quality training data has sparked legal challenges. A recent ruling deemed that training AI on legally acquired books can be considered fair use, a partial win for companies like Anthropic, Meta, and OpenAI. However, the legal landscape is complicated by issues surrounding pirated materials. Anthropic recently settled a $1.5 billion lawsuit over scanned pirated works, while Salesforce faces a class action lawsuit regarding similar allegations.