Project Panama exposes the destruction of print books to train artificial intelligence models
Listen to this article
Read by Anchor
An analytical post by Anna's Archive has revealed a growing trend among artificial intelligence companies to buy large quantities of used print books through commercial brokers, scan them, and permanently destroy the physical copies. These measures come in pursuit of pure, human-generated training data from before 2022, prior to the widespread proliferation of machine-generated text. The report highlighted disclosures from a 1.5 billion dollar copyright settlement regarding the top-secret "Project Panama," launched by Anthropic in early 2024, in which the company spent tens of millions of dollars to acquire and scan millions of print books to train its Claude model before destroying the volumes entirely.
Cost calculations and digital data monopolies
The practice of destroying books after digitising them rests on strict commercial and technical calculations driven by the model race, as rapid destructive scanning costs far less than non-destructive scanning that demands substantial time and delicate handling to preserve the paper and binding. Beyond cutting operational expenses, disposing of the physical copies helps reduce legal exposure and ensures competitors cannot scan and train on the same materials. This mechanism gives firms holding exclusive digital datasets the upper hand, locking human knowledge inside private corporate servers away from the public domain.
The clean data crisis and the decline of human text
This struggle is taking on critical dimensions given the shifts across digital spaces, with reported data showing that synthetic AI content accounts for more than half of newly published online material since the start of 2025. The phenomenon raises serious questions about the ability of researchers and the public to distinguish authentic human text in the future, as once models consume the last printed human records and the physical originals disappear, most of the digital sphere will be dominated by text produced by machine algorithms.
An initiative to protect print heritage from destruction
In response to these developments, the platform has issued an international call for volunteers to digitise books, articles, periodicals, and rare manuscripts from libraries and archives worldwide, preserving them within open collaborative libraries before they vanish from public circulation. The initiative offers incentives, including lifetime memberships and funding for large-scale scanning operations, in an effort to safeguard non-proprietary human knowledge and counter the trend turning written heritage into closed, proprietary assets held by major technology firms.