Amazon Is Destroying Rare Books to Feed AI Training Data
An investigation found that Amazon is buying up rare, out-of-print books, cutting off their spines, and scanning the text to train its AI language models. Such books are valuable because their content isn't available online and predates AI-generated text.
According to an investigation by 404 Media, Amazon has been purchasing rare books, removing their bindings, and scanning them for text to use in training its artificial intelligence models. Reporters tracked the process by planting a tracking device inside a rare book, which was ultimately traced to an Amazon facility in Las Vegas known as VGT3 — a warehouse identified by a logo depicting a dinosaur clutching a book.
In a statement to 404 Media, Amazon said it "purchases books through commercial channels to improve the products and services customers use," without directly confirming the books are being used for AI training.
Why rare books matter
Large technology companies need enormous volumes of text to train their large language models (LLMs), and material readily available online has largely been exhausted. Some companies, including Anthropic, have previously faced accusations of using illegally pirated book text for training. Rare and out-of-print books whose contents can't be found online have emerged as a new, sought-after source of training data.
These older texts hold particular value because anything published before 2022 is guaranteed not to have been written by an AI system. This distinction matters because training LLMs on too much AI-generated text can cause "model collapse," a phenomenon in which the quality of a model's outputs deteriorates after ingesting excessive amounts of machine-generated content.
