"Project Panama": Anthropic was caught secretly buying, scanning, and physically shredding millions of books to train Claude.
A federal judge recently unsealed over 4,000 pages of court documents from an ongoing copyright lawsuit, revealing a secret internal program at Anthropic codenamed "Project Panama."
Inside the unsealed files was a planning document with a sentence that sums up the stealth operation: "We do not want it to be known that we are working on this."
Here is the breakdown of how one of the leading "AI safety" companies allegedly built its training dataset behind closed doors:
Phase 1: Mass Piracy and a $1.5 Billion Settlement
Back in 2021, AI models desperately needed clean, high-quality text for training. Anthropic co-founder Ben Mann personally downloaded at least 5 million books from online shadow libraries/piracy sites.
Internal documents reveal CEO Dario Amodei wanted to bypass traditional licensing deals to avoid what he called the "legal slog." That shortcut eventually caught up with them, leading to a massive $1.5 billion settlement—the largest copyright payout in history.
Phase 2: Industrial Book Shredding as a Legal Strategy
After getting caught pirating digital copies, Anthropic pivoted to a bizarre physical workaround:
They began purchasing used physical books in bulk—tens of thousands at a time.
They ran the books through industrial hydraulic cutting machines to slice off the spines.
High-speed scanners digitized every page into training data.
The remaining paper was shipped directly off to recycling plants.
The Loophole: The Shredder IS the Defense
The most dystopian part of the unsealed documents? A federal judge ruled that this physical scanning-and-shredding pipeline was completely legal. Because Anthropic bought physical copies and destroyed the originals after scanning them, it bypassed standard copyright infringement under fair use loopholes.
The physical destruction of the books was literally the core legal defense.
Whether Anthropic was pirating digital files or physically slicing up paperbacks with hydraulic blades, the end result remains identical: authors received zero royalties for their work being used to train frontier LLMs.
What do you think of "Project Panama"? Is physical book destruction a clever legal loop-hole or a deeply dystopian look at how AI data is sourced?