AI companies are reportedly buying millions of secondhand books through intermediaries, scanning them for training data, and systematically destroying the physical copies. An investigation by 404 Media revealed a coordinated strategy to source high-quality, pre-2022 content while avoiding public scrutiny and copyright lawsuits.
The strategy emerged because AI training data faces a fundamental problem. The internet is saturated with “AI slop,” low-quality machine-generated content that contaminates training datasets. When AI learns from AI-generated text, output quality degrades. Books published before 2022 offer a solution. Print materials predate the generative AI explosion and contain dense, human-authored knowledge unlikely to be contaminated by synthetic content.
ISBNdb, the world’s largest book database, now markets this advantage directly to AI companies. The platform states that print books from the pre-LLM era are “structurally guaranteed to be free of this contamination.” ISBNdb explicitly tells clients that “physical books published before this date are structurally clean of modern poisoning tools.” More candidly, the platform acknowledges the reputational risk. “The optics problem is real,” ISBNdb’s website says. “AI company destroys two million books is not a headline that generates sympathy.”
Book purchasing accelerated sharply starting April 2026. ISBNdb now facilitates bulk orders ranging from 1,000 copies to one million books in single transactions. Anonymous booksellers report sales jumping from approximately 20 books weekly to several hundred.
Anthropic established the destructive precedent, as the company used hydraulic-powered cutting machines to remove pages from books. They then scanned them using industrial imaging equipment. A federal judge found this practice transformed the original texts into new digital forms, qualifying as fair use. Anthropic still paid $1.5 billion for maintaining pirated copies.
Booksellers in the Netherlands reported targeted bulk purchases of scarce and out-of-print titles. Some copies exist nowhere else globally. There’s a growing sense of unease rippling through the industry. Platforms like ISBNdb are making it easy for these bulk orders to happen, all while keeping the identities of the AI buyers under wraps, just as they promised. While there might be some strong clues suggesting that a bulk order is destined for an AI lab, booksellers can mostly only guess and wonder.
