According to reporting by 404 Media cited by Futurism, several companies working on artificial intelligence have been buying large quantities of older and often rare books, scanning them for training data and then destroying the physical copies.
Using books as training material for AI is not new: companies such as Meta and Anthropic have done so. Reports say Anthropic purchased used books, removed their spines and scanned the pages; after processing the physical volumes were apparently discarded.
Observers say the practice was enabled by a judicial interpretation that allows buyers to process the content of purchased books rather than resell it, which the buyers consider within their consumer rights. Encouraged by that legal reading, book acquisition firms that previously served libraries and bookstores are now trying to join the AI supply chain to profit from demand.
The metadata provider ISBNdb highlights on its website that, in its view, older books make better training material for AI than modern internet content or newer books. 404 Media reports that larger AI developers order several thousand, and in some cases up to one million, volumes from ISBNdb at a time. Because destroying books would create negative publicity for developers, ISBNdb also offers a confidentiality agreement alongside its service.
Why this matters: the practice raises questions about physical preservation and ethics. The disappearance of rare or unique copies can affect library collections, antiquarian markets and cultural heritage. At the same time, AI developers appear to be seeking texts written and edited before the widespread adoption of chatbots after 2022, in order to reduce the proportion of AI-generated material in their training sets.
The reporting indicates growing demand for older books and the emergence of intermediaries that collect or broker volumes specifically for AI training. Details about the full extent of the practice and the motivations of all parties involved remain under investigation, according to the media coverage.
