Destroying Books Saved Anthropic in Court: Inside the AI Physical Book Scanning Industry

APPSO · wechat · 2026-07-31

This article offers an in-depth look into Anthropic's massive acquisition and destructive scanning of physical books to train Claude.

Project Panama: Anthropic hired a former Google Books executive and spent tens of millions purchasing millions of physical books. To maximize digitization throughput, they destroyed the books by slicing off their spines and feeding the pages into high-speed scanners.
Court Rulings & 'Fair Use': In a copyright lawsuit, the judge ruled that Claude learning from books to generate new text is highly transformative 'fair use'. Dramatically, destroying the physical books became Anthropic's shield—the court viewed this as a 'one-to-one format conversion' that didn't increase the total number of copies.
Piracy Penalties: In contrast, Anthropic faced penalties for downloading over 7 million pirated e-books from sources like LibGen, eventually agreeing to a $1.5 billion settlement in 2026.
Emerging Gray Industry: This case has spawned a new industry targeting AI companies. Vendors now bulk-purchase pre-2022 physical books, marketing them as 'clean' training data unpolluted by AI, leading to suspicious bulk orders for out-of-print and niche books.

More from AGI Musings