Amazon Facility Scans and Destroys Books for AI Training

Employees at the facility reportedly receive massive shipments of books and then cut off the spines to allow the pages to be fed more quickly into scanners. This process inherently destroys the physical book. Amazon has confirmed that it "purchases books through commercial channels to help develop and improve the products and services our customers use," and stated that this specific operation had not been previously reported.

The demand for physical books for AI training is driven by the need for pre-2022 text that is not largely available online or machine-written. Training AI models on their own output risks a degradation in quality, a phenomenon known as "model collapse." Therefore, older, printed books are seen as "clean fuel" for AI development.

Amazon is not the only entity engaged in this practice. A secondary market has emerged to supply AI laboratories with physical books that are subsequently stripped, scanned, and discarded. Companies like Anthropic have reportedly digitized millions of books through similar destructive scanning methods as part of internal projects. This scramble for data has led to legal challenges, with a federal judge ruling that training AI on legally purchased books can be considered fair use, a decision that has been claimed by Meta and OpenAI as well. However, other companies like Anthropic have faced settlements over scanned pirated titles, and Salesforce is currently involved in a class-action lawsuit concerning alleged book piracy.