Inside 'Project Panama': How Anthropic Secretly Scanned and Shredded Millions of Books to Build Claude

Somewhere in a warehouse, a hydraulic cutting machine sliced the spine off a book. A production-grade scanner digitized every page in seconds. What was left of the physical book went straight to a recycling bin. Multiply that by millions, and you've got what court documents now call "Project Panama" — Anthropic's secret, years-long operation to buy, scan, and destroy an enormous share of the world's printed books to train Claude.

The story broke wide open this week after The Washington Post reviewed more than 4,000 pages of unsealed court filings, and it hasn't stopped spreading since — pulling in reactions from Elon Musk, investor Michael Burry, and secondhand booksellers across Europe who say they're suddenly getting mysterious bulk orders for rare titles.

Quick Summary & Key Takeaways

• What "Project Panama" Was: A 2024 Anthropic operation that bought millions of physical books, scanned every page, and destroyed the originals to build Claude's training data.
• The Method: Vendors used hydraulic cutting machines to remove book spines, then high-speed scanners to digitize pages, before recycling what remained.
• The Scale: One vendor proposal described capacity to convert between 500,000 and 2 million books over six months; the total spend ran into tens of millions of dollars.
• It Started With Piracy: Before buying books legally, Anthropic co-founder Ben Mann downloaded a 196,640-title pirated library in 2021, followed by millions more titles from other pirate sources.
• Legally, It's Split Down the Middle: A federal judge ruled that destructively scanning legally purchased books counts as fair use. The separately pirated books led to a $1.5 billion settlement — the largest copyright payout in US history.
• The Public Reaction Was Swift: Elon Musk said he's asked his SpaceX AI team to scan rare books non-destructively instead, while investor Michael Burry called the practice "evil incarnate."
• Booksellers Are Feeling It Now: Secondhand booksellers report a surge in anonymous bulk orders for rare, pre-2022 titles, believed to be fueled by AI training-data demand.

What's New in This Story

New Detail	What It Means
4,000+ pages of unsealed court filings	Turned a previously vague lawsuit detail into a fully documented, traceable operation
Judge ruled destructive scanning is fair use	Sets a legal precedent that could shape how every AI company sources book data going forward
Anonymous bulk book orders surging	Suggests the book-shredding pipeline for AI training didn't stop with Anthropic's case — it may be industry-wide now
ISBNdb named as a facilitator	A previously low-profile book database is now under public scrutiny for enabling anonymous bulk purchases
Elon Musk publicly breaking ranks	A rare instance of a rival AI company founder criticizing a specific competitor's data practices in public

Why Destroy the Books At All?

This is the part that confuses most people, and it's actually the crux of the legal case. Keeping both the physical book and a digital scan would have meant Anthropic held two copies of the same copyrighted work — a much harder position to defend in court. By destroying the physical original the moment it was digitized, the company could argue it had simply converted one legally purchased copy into a different format, not created a new one.

The presiding judge in the case largely accepted that logic, ruling that converting a purchased physical book into a searchable digital copy — and destroying the print original in the process — qualifies as fair use. It's a narrow, format-shifting argument, but it's now a legal precedent other AI companies sourcing training data from books will likely lean on.

The Part That Isn't Fair Use: The Piracy

The book-shredding operation is only half the story, and it's the more legally defensible half. Separately, Anthropic's court case revealed that the company's book-training pipeline began with piracy, not purchases. Co-founder Ben Mann reportedly downloaded a collection of nearly 200,000 pirated titles in early 2021, and millions more pirated books followed from other shadow libraries over the next year.

That piracy — not the legally purchased and destructively scanned books — is what led to Anthropic's $1.5 billion settlement with authors, working out to roughly $3,000 per affected book. It's the largest copyright settlement in US history, though relative to Anthropic's reported revenue run rate and valuation, several analysts have noted the payout amounts to a small fraction of the company's overall size.

Why This Is Spreading Far Beyond Tech Circles

What turned this from a legal footnote into a genuinely viral story is the physical, visceral detail — machines literally slicing the spines off books, pages fed through scanners, and the paper going straight to recycling. That image landed hard on social media, prompting reactions from people well outside the AI industry.

Secondhand booksellers across Europe have reported a wave of unusual bulk orders for rare, pre-2022 titles from anonymous buyers, and pointed to ISBNdb — a large book database — as a channel enabling these purchases while keeping buyers unidentified. Whether all of that demand traces back to AI training pipelines specifically isn't fully confirmed, but the timing and scale have been enough to fuel widespread suspicion that Anthropic's approach wasn't an isolated case.

💡 AI Tech Safar Insight
The legal outcome here is almost beside the point for how this story is landing publicly. A judge can rule that destroying a book to digitize it is technically fair use, and that ruling can still feel deeply uncomfortable to anyone who values physical books as more than raw training material. What's really driving the backlash isn't a copyright technicality — it's the image of irreplaceable print copies being fed through a blade for the sake of a training dataset, at a scale most readers never imagined was happening.

Frequently Asked Questions (FAQs)

Q1: What was Project Panama?
Anthropic's internal codename for a 2024 operation to buy millions of physical books, scan every page using industrial equipment, and destroy the originals to build Claude's training data.

Q2: Is destroying books to scan them actually legal?
For legally purchased books, yes — a federal judge ruled that converting a purchased print copy into a digital one, while destroying the print original, qualifies as fair use.

Q3: Then what was the $1.5 billion settlement for?
That settlement covered a separate issue — millions of pirated digital books Anthropic downloaded from shadow libraries, not the legally purchased books it scanned and destroyed.

Q4: Are other AI companies doing this too?
It's not fully confirmed, but secondhand booksellers report a recent surge in anonymous bulk orders for rare books, suggesting the practice may extend beyond Anthropic alone.

Q5: How did Elon Musk respond?
Musk said he has asked his SpaceX AI team to scan any rare books non-destructively, preserving the physical copies instead of cutting off the spines.

What Do You Think?
Does a legal "fair use" ruling settle the ethics of destroying physical books for AI training, or does it just mean the law hasn't caught up yet? Drop your take in the comments below!

Related Reading:

• Frontier AI Security 101: Sandboxes, Breaches & Risks
• OpenAI's Rogue AI Breached 4 Services, Left Notes for Itself — And Anthropic Faces Backlash
• 1,100+ Employees at OpenAI, Anthropic, Google & Meta Ask Washington to Prepare for an AI Slowdown

Source: Reporting based on IBTimes UK, Yahoo Finance, and MLQ News.