I cut up millions of books to make Claude: The truth about Anthropic’s Project Panama

by

$1.5 Billion Settlement, Millions Destroyed: Key Takeaways

  • Anthropic purchased and destroyed millions of books to train Claude, scanning them before disposal.
  • Internal document: “Project Panama is our effort to destructively scan the world’s books.”
  • $1.5 billion settlement, authors to receive approximately $3,000 per book.

What Happened?

Anthropic’s secret project was revealed through the release of over 4,000 pages of court documents. The codename was “Project Panama.” Internal planning documents stated, “Project Panama is our effort to destructively scan the world’s books.” They purchased tens of thousands of books in bulk from used bookstores like Better World Books and World of Books. They neatly cut off the spines with a “hydraulic guillotine.” They scanned the pages with high-speed, high-quality scanners. And then a recycling company collected the remains.[Techmeme]

The project was led by Tom Turvey, a former Google executive who created the Google Books project 20 years ago. For about a year, Anthropic invested tens of millions of dollars in acquiring and scanning millions of books.[Futurism]

Why is it Important?

Frankly, this shows the reality of acquiring AI training data.

Why did Anthropic choose this approach? First, to avoid the risk of illegal downloads. Second, purchasing used books and disposing of them as they wished was likely legal under the “first sale doctrine.” In fact, the judge recognized this scanning method itself as fair use.[CNBC]

However, there was a problem. Before Project Panama, Anthropic had freely downloaded over 7 million books from illegal sites such as Library Genesis and Pirate Library Mirror. The judge ruled that this part could constitute copyright infringement.[NPR]

Personally, I think this is key. The problem is that they illegally downloaded books first, rather than scanning them to destroy “legal” books. Anthropic themselves were aware of this. Internal documents stated, “We don’t want this work to be known.” Will it work?

The $1.5 billion settlement is the largest in the history of AI copyright disputes. Approximately $3,000 per book will go to the authors for the estimated 500,000 books.[PBS]

AI has another precedent. The impact on companies is significant. OpenAI, Google, and Meta are also facing similar lawsuits. The standard has become clear: “Buying and scanning books is okay, but illegal downloads are not allowed.”

Anthropic is already embroiled in a music copyright lawsuit. A separate lawsuit was filed in January, with music publishers claiming that Claude 4.5 was trained to “memorize” copyrighted works. Watchdog]

Frequently Asked Questions

Q: How many books were actually usable in Project Panama? Were they destroyed?

A: According to court documents, up to 2 million books were subject to “destructive scanning.” Anthropic purchased tens of thousands of books from used bookstores such as Better World Books and World of Books, and it is estimated that they processed millions of books over about a year, investing tens of millions of dollars.

Q: How much will the authors receive?

A: The $1.5 billion settlement applies to approximately 500,000 books. That’s about $3,000 per book. Authors of illegally downloaded books are eligible to claim, and can claim individually once the settlement is approved by the court. However, the actual amount received may increase if not all authors claim.

Q: Is it legal to buy and scan books?

A: The judge recognized this method as fair use. This is because books purchased under the “first sale doctrine” can be disposed of as desired. However, Anthropic’s problem is that they downloaded books from illegal sites before Project Panama. Scanning legally purchased books is now legal.

If you found this article helpful, please subscribe to AI Digester.

References

Leave a Comment