Amazon — the company that started life as an online bookstore — is buying rare books, cutting off their spines and scanning them to train AI models, according to a new investigation by 404 Media, which placed a tracking device in a rare book and followed it to an Amazon facility in Las Vegas.

The facility, known as VGT3 and identified by a symbol of a dinosaur holding a book, is where the books are processed. Amazon told 404 Media that it "purchases books through commercial channels to improve the products and services customers use." The company did not dispute that the books are being destroyed in the process.

Why rare books? Companies training large language models need unfathomably large amounts of text — and most of what is freely available online has already been ingested. Out-of-print and hard-to-find books offer a fresh source of coveted training data. Critically, nothing published before 2022 could possibly have been written by an LLM. When models train on AI-generated text, they risk "model collapse", a degradation in output quality that occurs after ingesting too much machine-written material. Pre-LLM books are therefore guaranteed to be free of what the industry calls AI slop.

The practice has been quietly growing. Court filings earlier this year revealed how Anthropic "destructively" scanned millions of books to build its models, and booksellers in Australia and beyond have raised the alarm about mysterious bulk orders that see old titles scanned, then destroyed. ISBNdb, a company that sources printed books for AI firms, markets the practice to clients — while acknowledging that "the optics problem is real."

For Amazon, the irony is thick: the retail and cloud giant that began as a bookseller is now part of a supply chain that pulps the printed word to feed the next generation of AI. The scale of Amazon's own involvement — how many books, from which sources — remains unclear, and the company has not commented on the environmental or cultural cost.

Sources