AI
Amazon Acquires and Destroys Rare Books to Feed AI Training Systems
The e-commerce giant is purchasing scarce volumes, removing their spines, and scanning them as training material for large language models.
TechCrunch — AI
Amazon is procuring rare books in bulk, destroying them for data extraction, and funneling the content into its AI training pipeline, according to investigative reporting by 404 Media, which secretly tracked a rare book shipment to an Amazon facility in Las Vegas. The company confirmed the practice, stating it "purchases books through commercial channels to improve the products and services customers use."
Why rare books matter for AI training#
The demand for textual input to train large language models has grown insatiable—tech companies have already exhausted what's freely available on the internet and turned to ethically murky alternatives, with some, like Anthropic, facing legal consequences for using pirated materials. Out-of-print and scarce books represent an untapped reservoir of original human-authored content.
The advantage of pre-digital text#
Books published before 2022 carry a critical advantage: they were certainly written by humans, not generated by AI systems. This distinction has become increasingly important as models risk degradation from consuming their own outputs. Researchers have identified a phenomenon called "model collapse"—a decline in output quality that occurs when language models train excessively on AI-generated text rather than original human writing.
