A bookseller had a suspicion. He watched bulk orders of rare books that made no commercial sense. Nobody buys a first edition of a niche academic text in quantities of fifty. So he planted an AirTag inside one of the books. He shipped it as part of a regular order. Then he watched the GPS signal travel straight to an Amazon AI training facility in Las Vegas.
Inside, workers tearing books from their spines and feeding pages into industrial scanners.
This has been an open secret in rare book circles for about a year. Booksellers noticed the pattern. Bulk purchases of titles no retail customer would ever want in those quantities. They suspected AI firms were behind it. Now there is a paper trail. Or rather, a GPS trail.
Amazon's official response was a masterclass in saying nothing. "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use." No mention of AI. No mention of destruction. No acknowledgment that rare, sometimes irreplaceable cultural artifacts are being shredded after a single scan.
The race to build frontier AI models is so resource-constrained that companies are destroying physical objects to stay competitive. They are not pirating or licensing the books. They are buying them and feeding them into a shredder after one pass. The training data arms race has real consequences. Most of them stay invisible until someone hides a tracker in a book.
Here is the ironic twist that ties this to the next story. The same industry that scans other people's work without permission is also rushing to watermark its own output. Anthropic, the company behind Claude, recently announced it will watermark everything its models touch. Invisible text watermarks. Signed provenance metadata. All to signal AI involvement. They called it the Scarlet Letter program.
So the people taking your work without asking want to make sure nobody mistakes their work for yours.