AI companies may be gobbling up old books, and I really hope they aren’t destroying them
Booksellers across several countries are seeing unusual bulk orders for old and out-of-print books, and some suspect AI companies may be behind the buying spree.
Secondhand booksellers have recently observed a peculiar pattern in bulk orders for obscure books. U.K. and Irish sellers have received orders for an array of subjects, from agriculture to racing biographies. Buyers often use anonymous aliases, send orders to the same consolidators, and pay full price without negotiation. Similar activity has emerged in the U.S., Australia, and Europe. While there's no evidence linking these purchases to AI companies, the situation raises concerns.
404 Media initially reported finding an ISBNdb landing page advertising "Printed Books Sourcing for Your AI LLMs Dataset Needs." The page disappeared on July 28, claiming it was part of an exploratory effort and that ISBNdb had moved away from the idea. Two days later, ISBNdb issued a statement denying any book purchases, scanning, or selling for AI training, nor had they launched the proposed service.
Speculation links Singapore-based AI data company 2077AI to requests for thousands of specialized books. The theory suggests that pre-AI boom books hold valuable human-written content with minimal risk of containing AI-generated text. However, destructive scanning practices have been documented. Anthropic, for instance, acquired physical books, cut off their spines, scanned the contents, and sent the books for recycling while building its training library.
As a book enthusiast, the idea of destroying rare or out-of-print books for AI training is troubling. Once those copies disappear, finding another can be difficult. Anthropic aims to make AI-generated text more identifiable by adding an invisible watermark to Claude-generated text. This approach changes how Claude selects words and creates a statistical pattern detectable later. Despite Anthropic's intentions, the prospect of AI companies destroying books for their training data is disconcerting.
Written by urgent.news from Digital Trends's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
