What happened
According to 404 Media, Amazon is buying large quantities of rare books, cutting off their spines and scanning the pages to produce training data for its AI models. The outlet hid a tracking device inside a rare book, and the book ended up at an Amazon facility in Las Vegas.
The site is called VGT3. Its logo is a dinosaur gripping a book. We did not make that up, and honestly we could not have.
Amazon's statement to 404 Media was short. Amazon's statement says it buys books through normal commercial channels to improve its products and services. It did not deny the practice, and collectors are left wondering how many rare copies end up in a scanner rather than on a shelf.
Why it matters
Language models are hungry, and the open internet is running out of leftovers. Per TechCrunch's summary, companies like Amazon need enormous amounts of text, and their LLMs have already absorbed what they could find online. In Anthropic's case, that reportedly included pirated books.
So the hunt moves offline. Rare and out-of-print books are hard to find on the internet, which makes them a fresh source of coveted training data.
There is also a quality angle. Anything published before 2022 cannot have been written by an LLM. That matters because training on AI-generated text risks "model collapse", where output quality degrades after the model ingests too much machine-written material.
Old paper is now a clean-data supply chain.
The uncomfortable part is the destruction. Cutting a spine is how you scan a book quickly, and the result is a rare object that is no longer a book. The reporting frames this as buying through normal commercial channels, so the debate is less about legality and more about what we are willing to shred for a model.
The story rests on one reliable report plus Amazon's brief statement, so details on scale and which titles are affected are still thin.
What to do with it
- Watch the data trail.When a vendor touts a new model, ask where the training text came from and whether anyone asked the owners.
- Value your archives.If you hold out-of-print or pre-2022 material, know that clean human-written text now has real appeal to AI builders.
- Follow the sourcing fight.Data provenance is becoming the main argument around AI training, so track how companies answer it.
- Think before you sell.If you deal in rare books, check what a buyer plans to do with them before the spine comes off.
Ciao, sapiens. Hug a paperback while it still has a spine.


