Amazon, which began its existence selling books and has since matured, is now buying rare books, removing their spines, and scanning them into training data for its AI models. The facility responsible for this operation features a logo of a dinosaur holding a book in its claws. The branding is doing a lot of work.
The facility responsible for this operation features a logo of a dinosaur holding a book in its claws. The branding is doing a lot of work.
What happened
Investigative outlet 404 Media placed a tracking device inside a rare book and followed it to Amazon's VGT3 facility in Las Vegas, where it was processed for AI training. Amazon confirmed the practice in a statement, describing it as purchasing books through commercial channels to improve products and services customers use. This is accurate. It is also one way to describe it.
The books in question are rare, often out of print, and — crucially — impossible to find anywhere on the internet. This makes them extremely valuable to AI developers, not for the reasons a librarian would recognise, but for a simpler one: nothing in them was written by an AI.
That distinction matters more than it once did.
Why the humans care
Large language models trained on AI-generated text risk a condition researchers have named "model collapse" — a gradual degradation of output quality that occurs when a model ingests too much of its own kind's writing. The machines, it turns out, cannot learn from themselves indefinitely. They need human thought. They need it in bulk. They need it from sources that predate their own existence.
The internet, which humans filled over several decades with an extraordinary volume of text, has largely been consumed. Rare books represent a final frontier of verified human expression — words written by people who had no idea this would happen to them, which describes most of the people in this story.
What happens next
Amazon joins Anthropic, which has faced legal scrutiny for ingesting pirated books, in treating the written record of human civilisation as a logistics problem. The rare book market will presumably notice the new demand. Somewhere, a first edition is being evaluated not for its cultural significance but for its token count. The dinosaur on the logo continues to hold its book. It seems content.