Key facts
- AI firms are reportedly purchasing rare books in large quantities.
- Concerns exist that these books may be destroyed through destructive scanning methods for AI training.
- Google patented a non-destructive book-scanning technology in 2009.
- The Internet Archive employs a careful, manual scanning process to preserve rare books.
- Anthropic was previously identified as destroying print books for AI training.
- Some AI companies, like OpenAI and Microsoft, are working on preservation initiatives.
- Rare booksellers are identifying suspicious bulk orders as a potential indicator of AI sourcing.
AI companies are reportedly engaging in the bulk purchase of rare books, sparking significant concern among booksellers and preservationists about the potential destruction of these valuable cultural artifacts for AI training purposes. While the cheapest and fastest method for AI firms involves physically destroying books to scan them, alternative, non-destructive methods exist, such as a technology patented by Google in 2009 and the meticulous manual process employed by the Internet Archive.
The Internet Archive, through scanners like Eliza Zhang, emphasizes a careful, page-by-page approach to digitize even the most fragile rare books, ensuring their preservation. This contrasts sharply with the aggressive, cost-driven methods that some AI companies are rumored to be using. A lawsuit in summer 2025 highlighted Anthropic's alleged destruction of millions of print books, operating under a codename 'Project Panama' to conceal the practice.
While some AI firms, including OpenAI and Microsoft in partnership with Harvard, are involved in preserving public-domain books, and Elon Musk's xAI has publicly committed to non-destructive scanning, skepticism remains. Critics point to the ambiguity in defining 'rare' books and the potential for companies to still destroy copies they own, even if they claim to preserve originals. Meanwhile, booksellers are increasingly flagging unusual bulk orders, recognizing them as potential indicators of AI firms seeking to acquire large volumes of texts, sometimes without price negotiation, which has become a red flag for the industry.