📣 Send us your press release
Site updates every 15 minutes
Technology

Amazon buys rare books for AI training, raising preservation concerns

Amazon is acquiring old and rare books for destructive scanning for AI training purposes. Booksellers and preservationists have voiced concerns that destroying physical copies could render certain titles irreplaceable.

19 August 2026
Amazon buys rare books for AI training, raising preservation concerns
Image is an AI-generated illustration

Amazon has been acquiring old, obscure, and rare books for destructive scanning for AI-related work, according to an investigation by 404 Media. Booksellers and preservationists have expressed concerns that the destruction of these physical copies could make certain books irreplaceable.

The investigation revealed that Amazon sends purchased books to a facility, VGT3 in Las Vegas, for a destructive scanning process. Workers remove book spines and separate pages for faster scanning, after which the physical copies are discarded. This practice is particularly concerning for old and rare books, which may exist in limited numbers or contain unique handwritten notes and other physical details that a digital file cannot preserve.

Amazon confirmed it purchases books through commercial channels to improve customer-facing products and services, but it did not specify the volume of its book purchases or whether it preserves rare titles. European booksellers have reported unusual bulk orders of thousands of obscure titles, suspecting AI companies or intermediaries are behind the purchases. These orders do not resemble typical collecting or library acquisitions.

Amazon is not the only major AI company turning to books. Previously, Anthropic, Meta, OpenAI, and Google have also utilized books for AI development. The distinctions lie in their methods: Amazon and Anthropic are documented to have acquired and dismantled physical books, while projects associated with OpenAI and Google have leveraged existing digitized collections. In the Anthropic case, a court ruled in 2025 that using copyrighted books for AI training constituted "fair use," though the defense was not accepted for books obtained through piracy.

Original source: medianama.com