Tech giants are buying and destroying thousands of rare and out-of-print books to train artificial intelligence models. Following a tracking experiment in August, investigations revealed that major platforms dismantle physical volumes at high-speed sorting warehouses in Nevada to secure pristine, un-digitized text data for advanced machine learning.
The five-hundred-year-old technology of printed books has collided with generative artificial intelligence in a destructive manner. While digital archives once promised permanent preservation, major technology companies are now dismantling physical volumes to feed data-hungry algorithms. The practice, brought to light through tracking experiments and investigative reporting, reveals that vast quantities of rare and out-of-print literature are routinely shredded after being scanned for machine learning systems.
The AirTag Tracking Experiment and the Nevada Warehouse
The scale of the operation became clear after an investigative team embedded an Apple AirTag tracking device inside a volume among a commercial order of rare books. The package was traced across the United States to a facility in Las Vegas known as building VGT3 (according to reporting by La Presse).
According to employee accounts shared on internal forums and detailed by media investigations, the processing method is mechanical and absolute. Workers routinely slice off the bindings to lay pages flat for high-speed scanners. Once digitized, the physical remains of the books are discarded and destroyed.
Automated Purchasing and the Hunt for Unique Data
The acquisition strategy relies on automated scripts capable of executing massive purchases at unusual hours. At a French second-hand book distributor known as Le Livre Vert, operational director Vera DaCunha noticed an abrupt surge in bulk orders (reported Franceinfo). An automated buyer acquired roughly 1,200 volumes totaling nearly 20,000 euros in a single month through a Canadian entity named Zoom Books.
The orders targeted diverse, out-of-print literature that normally attracts only niche collectors, including regional histories, specialized travelogues, and medical texts. The speed of the transactions precluded human involvement. As DaCunha observed, orders arrived in rapid sequence at 4:40, 4:41, and 4:42 in the morning (noted Franceinfo). In a parallel initiative dubbed Project Panama, artificial intelligence developer Anthropic pursued a similarly aggressive acquisition program to secure raw text (reported La Presse).
Why Tech Companies Target Physical Paper
The race to harvest printed volumes stems from a growing scarcity of fresh digital training material. Having ingested the vast majority of web-accessible text, artificial intelligence developers face diminishing returns from low-quality internet data and the legal hazards of pirated repositories. Un-digitized physical books offer pristine language patterns and verified structures.
Industry specialists emphasize that these physical texts provide a distinct advantage for machine learning architecture. Ari Kouts, an artificial intelligence specialist at Wavestone, explained that to compensate for the fact that nearly all existing digitized data has already been used, major AI model creators have gone looking for data that today exists only in the real world (cited by Franceinfo). By feeding algorithms text that has never existed online, developers expand the vocabulary and reasoning capabilities of their models.
Industry Backlash and the Legal Precedent
Publishing organizations have condemned the practice, pointing out the permanent loss of cultural heritage. Karine Vachon, general manager of the National Association of Book Publishers in Canada (ANEL), noted that treating rare physical volumes as raw pulp runs counter to the fundamental preservation of human knowledge (stated La Presse). While publishers warn of cultural erasure and copyright infringement, legal protections remain fractured across international jurisdictions.
In the United States, judicial rulings have provided initial legal backing to tech developers. A court ruled that purchasing and scanning millions of books for AI training was legal (reported La Presse), though related litigation led Anthropic to settle out of court for 1.5 billion US dollars (noted La Presse). When questioned directly about the operations at its facilities, Amazon issued a concise statement.
What Lies Ahead for Copyright and Cultural Preservation
As large technology firms continue scouring international markets for physical literature, legal challenges are expanding into new jurisdictions. Authors and publishing houses in Quebec have initiated multiple class-action lawsuits against major artificial intelligence developers (reported La Presse), with court hearings scheduled for the autumn. With no comprehensive domestic framework protecting Canadian authors from automated physical harvesting, industry groups are monitoring American legal outcomes as an indicator of future regulatory exposure.