Amazon, once a humble online bookstore, is now making headlines for a far less nostalgic practice: destroying rare books to fuel its AI training efforts. According to an investigation by 404 Media, the tech giant has been purchasing large quantities of rare books, slicing off their spines, and scanning their contents for machine learning purposes. The outlet even embedded a tracking device in a rare book, which ultimately arrived at an Amazon facility in Las Vegas.
That facility, codenamed VGT3, is marked with a peculiar symbol: a dinosaur clutching a book. In a statement to 404 Media, Amazon confirmed that it “purchases books through commercial channels to improve the products and services customers use,” though it did not elaborate on the specifics of its AI training pipeline.
The Insatiable Demand for Training Data
Amazon’s actions underscore a broader industry challenge: the insatiable appetite for text data to train large language models (LLMs). By 2026, most publicly available internet data has already been ingested by leading AI models. This has driven companies to seek out obscure, out-of-print, or otherwise scarce texts—like rare books—that remain largely absent from the digital sphere. These volumes offer a fresh, untapped reservoir of content, making them particularly coveted.
What makes these books especially valuable is their authenticity: anything published before 2022 predates the widespread use of LLMs, ensuring that the text is human-authored. This is critical because when LLMs train on AI-generated content, they risk "model collapse," a phenomenon where output quality degrades as the model ingests too much synthetic text. Rare books, therefore, represent a safeguard against this degeneration, offering a pure, human-written data source.
A Shifting Landscape: Legal and Ethical Questions
Amazon’s approach comes at a time when the legal landscape for AI training data is in flux. In 2026, a landmark settlement in the Anthropic copyright case (approved in July) has set precedents for how companies can use copyrighted material, but the destruction of physical books raises new ethical questions. Is it acceptable to destroy unique cultural artifacts to further AI development? While Amazon frames this as a commercial purchase, critics argue that the practice is short-sighted, prioritizing technological progress over preserving literary heritage.
As AI companies continue to expand their data horizons, the industry will need to address these trade-offs. For now, the image of a dinosaur clutching a book serves as a stark reminder of the lengths—and sacrifices—required to build the models of tomorrow.
via TechCrunch AI