AI Companies Destroy Rare Books After Scanning Them for Training Data

Artificial intelligence companies are increasingly purchasing old and rare books in bulk, only to destroy them after scanning their contents to train AI systems. For developers of large language models (LLMs), these physical books are viewed merely as raw material, with their content extracted and the original copies subsequently discarded. This marks a shift from previous practices where companies often relied on digital books or pirated copies.

In a settled lawsuit involving the prominent AI firm Anthropic, it was revealed that the company used a guillotine paper cutter to slice the bindings off purchased books so the pages could be fed through high-speed scanners. The text is then used to train their AI systems. This practice exploits legal provisions that allow purchasers to freely use goods they have bought, treating the destructive scanning as a transformative use that does not infringe copyright.

Services like ISBNdb, which claims to have the world’s largest book database, are now facilitating these bulk purchases. The company argues that print books published before 2022 contain cleaner, more reliable data than internet text, which is increasingly contaminated by AI-generated content. It guarantees anonymity to its buyers to shield them from public criticism.

One book dealer noted a dramatic surge in sales, from dozens to hundreds of copies per week, with orders often identified only by book ID numbers, strongly suggesting AI labs are the buyers. While the financial boon is welcome, the dealer expressed sadness that rare books, of which few copies remain, are being systematically destroyed and rendered unreadable for future human readers. The trend has ignited a fierce ethical debate over the destruction of cultural and scientific heritage to fuel technological advancement.