Inside Project Panama, Anthropic's Secret Effort To Scan and Shred the World's Books

Anthropic spent tens of millions of dollars secretly buying, scanning and destroying millions of physical books to train its Claude chatbot, according to more than 4,000 pages of unsealed court documents reviewed by the Washington Post.

The operation, internally named 'Project Panama', ran from early 2024 and involved hydraulic cutting machines slicing the spines off books before high-speed scanners digitised every page. The paper was then recycled, leaving no physical copy behind.

The Pirate Roots of Anthropic's Claude Training Data

Before turning to legal purchases, Anthropic co-founder Ben Mann personally downloaded millions of pirated books from LibGen and Books3 in 2021, according to Judge William Alsup's June 2025 order in Bartz v. Anthropic.

The court order found Anthropic pirated more than seven million book copies in total. Judge Alsup wrote that the company had legitimate ways to obtain the material but chose piracy instead, to avoid what CEO Dario Amodei called 'legal/practice/business slog'.

From Piracy to Purchase, the $1.5 Billion Pivot

That piracy exposed Anthropic to serious legal risk. In February 2024, the company hired Tom Turvey, the former head of partnerships for Google Books, to find what internal documents described as a way to obtain 'all the books in the world'.

An internal planning document stated plainly that 'Project Panama is our effort to destructively scan all the books in the world'. The same document instructed staff to keep quiet, adding that 'we don't want it to be known that we are working on this'.

Vendor proposals referenced converting between 500,000 and two million books over a six-month period.

Judge Alsup's Fair Use Ruling and Its Limits

Judge Alsup ruled that digitising legally purchased books was fair use. His reasoning centred on the digitisation being a straightforward format change, since Anthropic had simply replaced the print copies it purchased with more convenient, searchable digital versions, without creating new copies or redistributing them.

That ruling effectively gave legal cover to the destroy-after-scan model. Anthropic separately settled the piracy portion of the case for $1.5 billion in August 2025, without admitting wrongdoing.

Print Books Remain in High Demand Among AI Firms

The appetite for print books has not cooled. ISBNdb, a firm that brokers high-volume book acquisitions, now facilitates orders of up to a million books for AI clients and offers non-disclosure agreements as part of the service.

The company's own website states that pre-2022 books are attractive specifically because they predate widespread AI-generated text, arguing that 'books represent curated, peer-reviewed, domain-specific human knowledge, structured in a way no web crawl can replicate'.

Booksellers, speaking on condition of anonymity due to commercial sensitivity, have told the Washington Post that rare, out-of-print titles with few surviving copies are among the stock being bought and processed.

For consumers, the practical impact centres on what happens to secondhand and rare book markets as institutional buyers with far greater purchasing power enter the field. Sudden bulk orders can affect availability and pricing for collectors, students and casual readers alike, and once a rare or out-of-print title is destroyed after scanning, no replacement copy exists.

The case also sets a legal marker other AI firms are likely to watch closely, since Judge Alsup's fair use finding on destructive scanning, though specific to Anthropic's case, is already shaping how the wider industry approaches sourcing training data from print books.

For now, the battle over training data has moved from courtrooms to book warehouses. Judge Alsup's ruling offers AI firms a legal roadmap, but it also raises a question for the rest of us: what do we lose when the last physical copy of a book is fed into a scanner and never seen again?