AI Companies Scanning Rare Books for Training Data
An investigation by 404 Media has uncovered a disturbing practice in the world of artificial intelligence (AI) training. Amazon and Anthropic, two major players in AI development, have been buying up printed books in bulk, scanning them for data, and then destroying the originals.
The investigation used an AirTag to track a shipment of rare books from Amazon's warehouse in Las Vegas, where they were cut off at the spine and digitized. The scanned data is used to train Amazon's Nova models.
Amazon claims it buys books through commercial channels to improve its products, but book sellers suspect that AI companies are systematically scanning every book by ISBN number. Printed texts are particularly valuable because they often don't exist online and predate 2022, making them free of AI-generated content.