AI Companies Destroying Rare Books to Train LLMs
AI companies are buying and destroying physical books to train large language models, using hydraulic cutters to remove pages for scanning. Anthropic paid a $1.5 billion settlement for using pirated digital books, but also used first-sale doctrine and fair use to legally destroy physical copies. ISBNdb, a book database service, now facilitates bulk purchases of 1,000 to 1 million books per order for AI labs, promising anonymity. Small booksellers report sudden bulk sales of rare and out-of-print books, likely to AI companies. Rare booksellers in the Netherlands also suspect AI buyers. The practice destroys unique copies, raising concerns about cultural heritage loss.
Key facts
- Anthropic used a hydraulic cutting machine to remove pages from physical books for scanning.
- A judge found Anthropic's digitization of physical books 'transformative' and protected by fair use.
- ISBNdb helps AI companies bulk-buy 1,000 to 1 million books per order.
- ISBNdb promises to keep AI buyers' purchases anonymous.
- One bookseller went from selling 20 books a week to hundreds in April, likely to AI labs.
- Rare booksellers in the Netherlands report bulk purchases believed to be from AI companies.
- Physical books from pre-2022 are considered 'structurally clean' of AI-generated content.
- ISBNdb acknowledges the 'optics problem' of AI companies destroying books.
Entities
Institutions
- Anthropic
- Meta
- ISBNdb
- 404 Media
- Futurism
Locations
- Netherlands