Printed books are being scanned and destroyed to train AI

A new commercial market is rapidly developing around the acquisition and processing of printed books for artificial intelligence training. This practice involves buying books in bulk, scanning them into machine-readable formats, and in some instances, destroying the physical copies to create digital files for AI systems. This trend is moving beyond speculation, with court records and investigative reports detailing the emergence of this specialized industry.

Companies like ISBNdb are expanding their services to help clients procure large quantities of books. Investigations reveal that bulk purchases can range from around 1,000 to one million books, often with buyers' identities protected by strict confidentiality agreements. Many of these books are then sent to specialized firms for digitization. Some employ destructive scanning methods, such as cutting off the book's spine for faster, cheaper page scanning, though not all bulk purchases necessarily result in physical destruction.

“

the optics problem is real

Books published before the widespread adoption of generative AI, particularly before 2022, are reportedly in high demand. They are considered valuable sources of human-authored text, predating the proliferation of AI-generated content online. ISBNdb has acknowledged the sensitive nature of this practice, noting the "optics problem" and the potential for negative headlines.

The legal landscape for this activity has been shaped by cases like Bartz v Anthropic. In this copyright dispute, authors alleged that Anthropic copied their books without permission to train its AI system, Claude. A US District Judge ruled that using legally obtained books for AI training constitutes an "exceedingly transformative" use under copyright law. The ruling also stated that Anthropic could lawfully convert purchased print books into digital copies, even if the physical copies were destroyed, provided the digital files were retained internally. However, this ruling did not cover Anthropic's use of books from unauthorized sources like shadow libraries.

“

'AI company destroys two million books' is not a headline that generates sympathy.