Amazon Destroys Rare Books to Train AI Models at VGT3
Amazon is buying rare books, cutting off their spines and scanning them at VGT3 in Las Vegas to obtain pre-2022 text unavailable online for AI model training.
Summary
On August 17, 2026, TechCrunch reported that Amazon, which began as an online bookseller, is buying large quantities of rare books, removing their spines and scanning them for AI training. 404 Media tracked one rare book to Amazon’s VGT3 facility in Las Vegas, whose identifying symbol is a dinosaur holding a book. Amazon said it purchases books through commercial channels to improve customer products and services.
Large language model developers have consumed much of the usable internet, while TechCrunch notes that Anthropic used illegally pirated books. Rare, out of print texts unavailable online offer scarce training material, particularly because works published before 2022 could not have been written by an LLM. Such human-written data may help models avoid model collapse, in which outputs deteriorate after excessive training on AI-generated text.
Positives
- Pre-2022 books provide human-written material uncontaminated by large language model output.
- Rare and out of print works can supply training data unavailable on the internet.
- Amazon says it purchases the books through commercial channels to improve customer products and services.
Risks & concerns
- Amazon is destroying rare books by cutting off their spines before scanning them.
- 404 Media’s tracking device led from a rare book to Amazon’s VGT3 facility in Las Vegas.
- Large language models risk model collapse when excessive AI-generated training data degrades their outputs.
- TechCrunch notes that Anthropic used illegally pirated books for AI training.