ISBNdb Funnels Rare Books to AI Labs That Scan and Shred Them
TL;DR
- AI firms are bulk-buying antique books through ISBNdb at 1,000 to one million per order, then destroying the physical copies after scanning.
- Anthropic reportedly used a hydraulic-powered cutting machine to slice pages for scanning and paid a $1.5 billion settlement to authors.
- Pre-2022 print is prized because ISBNdb pitches it as structurally guaranteed to be free of AI-generated contamination.
Bulk buyers operating under strict NDAs are hoovering up antique and out-of-print books, extracting the text, then destroying the paper. That is the through-line of a Futurism report on how AI training pipelines have quietly begun scraping the physical world for text that predates the current flood of AI-generated web content.
The most concrete named example is Anthropic, which reportedly used a hydraulic-powered cutting machine to slice pages from books it bought from resellers, then scanned them on industrial-grade imaging equipment. A judge ruled the destruction transformative and covered by fair use; Anthropic was still forced to pay a $1.5 billion settlement to authors over the sourcing of some of those books. The bigger structural story sits with ISBNdb, which markets its book database as a matchmaking service between AI buyers and rare-book sellers, facilitating orders "anywhere between 1,000 to one million books per order" while keeping the buyers anonymous under nondisclosure.
The pitch to AI labs is that print books from the pre-LLM era, roughly pre-2022, are, in ISBNdb's phrasing, "structurally guaranteed to be free of this contamination" from synthetic text. That is the demand signal driving the volume. One bookseller told Futurism the trade had jumped from no more than 20 books a week to hundreds, with rare and out-of-print editions moving in the mix. ISBNdb itself concedes, in the same reporting, that "'AI company destroys two million books' is not a headline that generates sympathy."
The honest caveat is what the reporting does not give you. It does not name buyers beyond Anthropic, does not quantify how many destroyed volumes are actually last-copy rarities versus common used stock, and does not say what happens to the pulped remains. The judge's transformative finding is real, but its reach beyond Anthropic's specific facts is untested.
What is worth watching is whether libraries, universities, and national archives start pitching non-destructive licensing of their clean pre-2022 corpora to the same buyers. There is a real business here for anyone who can deliver provenance without the shredder.
Shared on Bluesky by 2 AI experts
-
Ida Momennejad @neuroai.bsky.social: “It was a pleasure to burn.” - Fahrenheit 451 futurism.com/artificial-i... →
-
futurism.com/artificial-i...
View on Bluesky →
Originally reported by futurism.com
Read the original article →Original headline: AI Companies Are Buying Antique Books, Ingesting Their Contents to Train Models, and Then Destroying Them at Incredible Scale, Even If Almost No Copies Remain