Project Panama: A Fahrenheit 451 for the age of AI

In Ray Bradbury's Fahrenheit 451, society has become deeply anti-intellectual, with critical thought, independent ideas and reading outlawed. The "firemen" no longer put out fires; they start them, burning books and the houses that contain them. Guy Montag, one of these firemen, gradually discovers what books contain: memory, history, contradiction and the possibility of thinking differently.

More than seven decades later, the real world has produced a version of Bradbury's nightmare that is less theatrical but, in some ways, more unsettling.

Some of the world's most powerful big tech companies are acquiring physical books, cutting them apart, digitising them and discarding what remains. The industrial-scale process, known as "destructive scanning", treats physical books as sources of data that can be extracted and digitised before the original objects are discarded.

The anatomy of Project Panama

The scale of this operation emerged through the federal copyright lawsuit Bartz v Anthropic PBC in the US District Court for the Northern District of California. Unsealed court filings in late July 2026 detailed a book-ingestion operation operated by Anthropic under the codename "Project Panama".

According to an internal planning memo from April 2024, the project was described as "our effort to destructively scan all the books in the world".

Anthropic hired Tom Turvey, the former head of partnerships for Google Books, to help execute the strategy.

The company's move towards physical books was partly shaped by legal considerations.

After settling a $1.5 billion copyright lawsuit over its use of pirated digital libraries, including the Books3 dataset, Anthropic's founders sought another way to acquire books. 

The first-sale doctrine offered one possible route: Once a person legally purchases a physical book, they can generally resell, alter or destroy that particular copy without obtaining permission from the copyright holder.

The approach was straightforward: buy the books, cut them open, scan the pages and dispose of the physical copies.

The scale was substantial. 

Vendor proposals disclosed in the litigation contemplated processing between 500,000 and two million books within a six-month period.

By late 2024, millions of volumes had reportedly been processed through hydraulic cutting machines, with their pages fed into automated imaging systems and the remaining material sent for disposal or recycling.

Unlike the book-burning depicted in Fahrenheit 451, the process involves mechanically separating the pages from the physical books before digitising their contents. The resulting digital files can then be retained while the physical copies are discarded.

The AirTag trail and the Vegas shredder

A parallel investigation provided a glimpse into how industrialised the process had become.

Independent booksellers using the online marketplace Biblio noticed large, automated and price-insensitive purchases of their stock. One seller, working with 404 Media, placed an Apple AirTag inside a 1,000-book order. The device travelled across the US before reaching VGT3, a secured Amazon logistics facility in Las Vegas.

Workers at the facility described a factory-style book-processing operation. Crates arrived, barcodes were scanned and books were taken to cutting stations, where industrial paper cutters removed their spines. The pages were then digitised, while the remaining material was discarded.

The VGT3 scanning team's internal mascot reportedly depicts an AI-generated Tyrannosaurus rex holding a book in its claws.

Amazon confirmed that it purchases books to "develop and improve" its products.

The image of Bradbury's fireman therefore has a modern counterpart. The flamethrower has been replaced by machinery designed to separate the contents of books from their physical form.

The shadow supply chain

The companies involved do not necessarily need to purchase books directly. Brokers and bulk aggregators can perform that role.

ISBNdb, which describes itself as the world's largest book database, has positioned itself as a bulk procurement broker for AI developers. It advertises contracts ranging from 1,000 to one million books, with non-disclosure agreements protecting clients' identities.

The buying spree has also spread to independent bookstores.

Australian booksellers have reported large, price-insensitive orders from Zoom Books, a Canadian book recycler, including seemingly random titles such as 1970s soil mechanics manuals and local histories.

In Britain, Barter Books received an order from Zoom Books equivalent to an entire week's sales — between 2,000 and 3,000 books. In Ireland, bookseller Tomás Kenny of Kennys Bookshop flagged an order for 5,000 obscure titles with no apparent attempt to negotiate the price.

The books were reportedly traced to a warehouse complex near London's Heathrow Airport, from where volumes could be routed to overseas scanning centres.

For booksellers, the transactions can be financially welcome. But they also raise questions about what happens to books that have been preserved for years when they are purchased for digitisation rather than reading or collecting.

Why destroy what you can simply digitise?

One reason is the AI industry's growing demand for "clean" training data.

The internet is becoming saturated with AI-generated articles, summaries, images and other synthetic content. Training new models on material produced by earlier models can create a recursive feedback loop known as model collapse, in which errors accumulate and linguistic diversity and semantic nuance deteriorate.

Older books offer an alternative. Printed works published before the consumer LLM era are less likely to contain AI-generated material. More importantly, they represent some of the most carefully edited, coherent and sustained examples of human language available.

To AI companies, the pre-2022 library is therefore a vast record of human thought.

But there is a fundamental difference between preserving a book digitally and destroying the physical object after digitising it.

A physical book is not simply a container of words. It is an independent reference point, carrying its edition, publication date, publisher, annotations, marginalia and ownership marks. Its physical existence does not depend on the continued goodwill of a corporation.

Once that physical object disappears, the digitised version becomes the surviving representation. Whoever controls that representation has significant influence over how the text is accessed, interpreted, summarised and reproduced.

This is becoming less and less theoretical.

AI platforms can rewrite established works, while researchers are experimenting with models capable of altering literary classics. Without a physical reference point, the distinction between preservation and revision becomes increasingly difficult to maintain.

That is where Fahrenheit 451 becomes more than a literary warning.

Bradbury imagined a society in which knowledge disappears because people have been persuaded that books are inconvenient, dangerous or irrelevant. The modern version may be more subtle. The argument is not that books must disappear, but that nothing is lost if their contents are preserved digitally.

Yet digitisation is not necessarily preservation.

A book that survives only as data controlled by someone else is not quite the same thing as a book that survives on a shelf.

In Bradbury's novel, Montag eventually finds people who preserve books by memorising them after their physical copies have been destroyed. Bradbury writes of the need for "the saving and putting away" to begin again, whether "in books, in records, in people's heads".

Today, the threat comes without the fire. Books can be digitised while their physical copies are discarded, leaving digital records as the surviving versions of the texts.

The result could be a world surrounded by information but increasingly separated from the physical records through which that knowledge was preserved.