Amazon buys rare books for AI training, reportedly destroys originals

Amazon has been purchasing large volumes of rare and out-of-print books, scanning them to feed its AI models, and then destroying the physical copies. An investigation by 404 Media, published on August 17, 2026, used an AirTag to trace an approximately 1,000-book shipment from a California bookseller directly to Amazon’s VGT3 facility in Las Vegas, where workers reportedly slice off book bindings to speed up the scanning process. The originals don’t survive the procedure. Once the pages are digitized, the physical books are discarded. Why old books, and why destroy them Amazon and other AI developers are specifically targeting books printed before 2022 because they contain text written entirely by humans. Post-2022 publications carry a growing risk of containing machine-generated content, which creates a garbage-in-garbage-out problem for training large language models. Cutting bindings is simply faster than carefully opening a book flat enough for a scanner to capture every page. Amazon has acknowledged that it acquires books for its “technology efforts” but has not disclosed how many books it has purchased or what specific AI applications the scanned data supports. A broader indu...