Home / Technology / Physical Books Destroyed for AI Training Data
Physical Books Destroyed for AI Training Data
4 Aug
Summary
- AI company reportedly dismantled physical books for faster scanning.
- Anthropic's 'Project Panama' involved millions of bulk-purchased books.
- A lawsuit resulted in a $1.5 billion settlement for authors.
AI companies are facing scrutiny for their data acquisition methods, with Anthropic reportedly dismantling physical books to accelerate AI model training. This practice, internally codenamed 'Project Panama,' involved acquiring millions of books in bulk and cutting off their spines for high-speed industrial scanning to convert pages into machine-readable text.
The aggressive approach aimed to create vast datasets for training AI models like Anthropic's Claude. This method of destroying original literature led to a class-action lawsuit from authors, resulting in a $1.5 billion settlement last summer. The court ruled that while training AI on books can be considered fair use, Anthropic infringed copyrights by retaining millions of pirated books.
While companies like OpenAI, Google, and Meta have encountered legal challenges regarding book data usage, Anthropic is the only entity publicly identified for physically destroying books for training purposes. A service called ISBNdb was also implicated in sourcing large quantities of books for AI companies, though it denies direct involvement in AI training data acquisition.
