"Millions of Rare Books Tracked to Shocking End... AI Companies' Chilling 'Panama Project'"

Anthropic and Amazon hire outside firms for book orders
To avoid copyright litigation and obtain pure text
Book spines cut off to increase scanning efficiency

It has been revealed that global big tech companies such as Amazon and Anthropic have been purchasing rare books and out-of-print volumes in bulk from the secondhand book market, using them to train AI models, and then destroying the original copies.

Image created using artificial intelligence (AI) to aid in understanding the article. ChatGPT

According to Forbes on August 17 local time, U.S.-based IT media outlet 404 Media reported that big tech firms are operating dedicated large-scale facilities where they order paper books in massive quantities, dismantle the bindings, scan the pages at high speed, and then dispose of the originals.

Previously, during a copyright infringement lawsuit filed by writers against AI developer Anthropic, an internal planning document called the 'Panama Project' was made public, bringing these suspicions to light.

According to the disclosed document, Anthropic purchased millions of physical paper books and outsourced the process of slicing off their spines. The pages were then fed into high-speed scanning equipment for digitization, after which the original copies were immediately shredded. It is reported that these shredded books included not only bestsellers, but also rare editions, antique books, and out-of-print volumes with significant preservation value.

Tracking Books with AirTags Reveals Final Destination: Amazon Warehouse

There have long been suspicions in the secondhand book market that AI companies have been secretly orchestrating bulk purchases, with claims of a "sharp rise in large-quantity orders for paper books published before 2022."

To verify these suspicions, 404 Media, the U.S. IT media outlet, conducted direct investigative tracking. Working with an anonymous rare book vendor who had received an order for about 1,000 paper books, they placed Apple location tracking devices (AirTags) inside the books to follow their journey.

The shipment departed from California and, after passing through Milwaukee and Colorado Springs, ultimately arrived at a dedicated Amazon logistics warehouse. On the entrance was a logo featuring a Tyrannosaurus about to devour an open book, and it was discovered that staff inside were assigned to scanning paper books throughout the day.

Avoiding Legal Risks and Securing 'Pure Data'

There are two main reasons why AI companies are investing significant resources to purchase and destroy physical paper books: to avoid legal liability and to secure uncontaminated, high-quality training data.

If companies were to illegally utilize so-called 'shadow libraries'—web archives of pirated content—they would risk astronomical copyright lawsuits from authors, publishers, and media organizations. By contrast, when physical paper books are lawfully purchased, the buyer can exercise ownership under the 'first sale doctrine' of U.S. copyright law, meaning that even if the originals are shredded, the company can avoid legal penalties. Furthermore, transforming the purchased print editions into digital files for internal model training could potentially be recognized as 'transformative fair use.'

From a data quality perspective, paper books are considered an irreplaceable resource. Information currently circulating on the Internet is contaminated with low-quality AI-generated text, which can degrade the performance of subsequent AI training. On the other hand, paper books published before 2022 were authored by humans and thoroughly edited, making them premium learning materials with logical and structured reasoning.

Strong Backlash from Bibliophiles... Concerns About the Extinction of Rare Cultural Artifacts

The exact number of books destroyed through these practices is unknown. However, Anthropic's internal documents stated a plan to "destructively scan at least 500,000 to as many as 2 million books under a single outsourcing contract," and confidential bulk orders reportedly range from 1,000 up to 1 million books per transaction. Secondhand bookstores in the United States, United Kingdom, and Europe report that "anonymous transactions for the purchase of hundreds or thousands of non-fiction and out-of-print books have become routine."

Particularly notable is Anthropic's internal document, which stated, "The Panama Project aims to scan every book in the world" and further expressed a desire for "this operation to remain secret and not be disclosed externally," fueling the controversy.

Once these facts became known, both AI skeptics and bibliophiles strongly condemned such actions, branding them as 'the embodiment of evil.' Forbes pointed out that the destruction of rare and out-of-print volumes—sometimes among the last remaining copies—means the loss of the books' historical value, which is a critical issue.

This content was produced with the assistance of AI translation services.