Tech Firms Buy Rare Books, Cut Spines to Scan for AI Training

Tech companies are purchasing large quantities of rare old books, scanning them and often destroying them in the process, in a race to find fresh material to train artificial intelligence models, according to an investigation by 404 Media.

Reporter Emanuel Maiberg told NPR's Juana Summers that a bookseller who had been receiving unusually large orders reached out to him, prompting the outlet to plant a tracking device inside one such shipment. The device led investigators to an Amazon warehouse.

Maiberg said the pattern was easy to spot: unlike typical buyers, such as university libraries seeking books on a specific subject, these bulk purchasers ordered hundreds of books with no coherent theme — a telltale sign the material was being harvested rather than studied.

A lawsuit from book authors against the AI company Anthropic previously revealed that firms have cut the spines off books to feed pages quickly through scanners.

Maiberg explained that AI companies now favor older, rare books because they have largely exhausted internet text as training material. Attempts to train AI systems on AI-generated text caused a degradation problem known as "model collapse," pushing companies to seek fresh, human-written material not readily available online.

He cautioned that "rare books" targeted in these orders are typically not prized first editions but obscure, low-circulation items — self-published works, niche instructional manuals or vanity-press books that have little current emotional or market value, even if their future worth is unknown.

Amazon confirmed to 404 Media that the warehouse identified in the reporting, VGT3, is used for AI training data, saying in a statement: "Amazon purchases books through commercial channels to help develop and improve the products and services our customers use." Maiberg said he believes AI training is one purpose of the facility, though it may not have been its original function.