AI Firms Buy Up to 1 Million Books to Scan and Destroy for LLM Training
Updated
Updated · Futurism · Jul 25
AI Firms Buy Up to 1 Million Books to Scan and Destroy for LLM Training
2 articles · Updated · Futurism · Jul 25
Summary
ISBNdb is now helping AI companies place bulk orders of 1,000 to 1 million physical books, reflecting a widening trade in buying, scanning and destroying print titles for model training.
Pre-2022 print books are being marketed as “clean” data untouched by AI-generated web content, giving labs a legal and practical alternative to disputed digital and pirated text sources.
Anthropic’s settled lawsuit detailed the process: used books were stripped with a hydraulic cutting machine and scanned industrially, a practice a judge said could qualify as transformative fair use under first-sale doctrine.
Booksellers say demand has jumped from about 20 books a week to hundreds, with random ISBN-heavy orders suggesting AI buyers and raising fears that rare or out-of-print copies are being pulped.
The trade is spreading beyond the U.S., with Dutch rare-book dealers also reporting suspicious bulk purchases as intermediaries promise buyer anonymity to avoid the backlash seen by Anthropic and Meta.