Reported by 4 sources

The short version

  • 404 Media used an AirTag to track a rare book to an Amazon AI training facility in Las Vegas.
  • Workers at the facility are reported to tear books from their spines and scan pages before destroying them.
  • The practice has drawn criticism from booksellers and critics who view it as destructive to cultural heritage.

An investigation by 404 Media has revealed that Amazon is systematically destroying rare books to train its artificial intelligence models. The report details how a tracking device, specifically an AirTag, was hidden inside a rare book purchased in bulk. The tracker led investigators to an Amazon AI training facility in Las Vegas, identified as warehouse VGT3. There, the book was processed by a team dedicated to tearing texts from their spines and scanning their pages before discarding the physical copies.

The discovery provides concrete evidence of a practice that booksellers have long suspected but struggled to prove. For months, industry insiders had noted unusual bulk orders of rare and antique books by tech firms. These purchases were often followed by the disappearance of the texts from circulation. The AirTag footage confirms that at least one major player, Amazon, is engaging in this destructive method to acquire unique training data for its frontier AI models.

News Journal

The facility in Las Vegas appears to operate with a singular focus on data extraction. Workers reported that the process involves dismantling books to access their content, regardless of the physical item's value or rarity. A logo on the warehouse door depicting a Tyrannosaurus rex devouring a book was documented by investigators, suggesting a tone-deaf approach to the backlash surrounding these activities. This imagery contrasts sharply with the preservation ethos typically associated with rare book collections.

Amazon has declined to comment specifically on the findings, offering only a generic statement about purchasing books through commercial channels to improve customer products and services. The company did not explicitly mention AI training in its response. However, the context of the investigation and the nature of the facility strongly suggest that the primary purpose is data acquisition for machine learning algorithms. This ambiguity has fueled further criticism from observers.

The motivation behind this strategy appears to be the need for high-quality, unique text data. As AI models become more sophisticated, the demand for diverse and rare training materials increases. Competitors like Anthropic and xAI have publicly stated they do not train on rare or antique books, potentially giving Amazon a competitive advantage by accessing these exclusive sources. The scarcity of such texts makes them valuable assets in the race to develop superior AI systems.

Reports from online forums indicate that workers at the VGT3 facility faced periods of shortage, with concerns that the warehouse might shut down if new supplies were not secured. This reliance on a continuous stream of rare books highlights the intensity of Amazon's data hunger. The fact that the facility remains operational despite these challenges underscores the company's commitment to this method of data collection.

The ethical implications of destroying cultural artifacts for technological gain have sparked outrage among book lovers and preservationists. Critics, including investor Michael Burry, have labeled the practice as 'evil incarnate.' The destruction of rare books represents a loss of historical and literary heritage that cannot be recovered. While digital scans preserve the text, the physical objects themselves are lost forever, raising questions about the value placed on material culture in the digital age.

Conversely, some workers have described the job as a 'nice' opportunity with flexible hours, suggesting a disconnect between the moral concerns of outsiders and the practical realities of employment. This perspective adds complexity to the narrative, highlighting how economic incentives can drive participation in controversial practices. However, this does not mitigate the broader impact on the rare book market and cultural preservation efforts.

The investigation also sheds light on why certain books are targeted over others. AI firms appear to be selective in their acquisitions, focusing on texts that offer unique linguistic patterns or historical insights not available in more common sources. This selectivity drives up prices and reduces availability for collectors and libraries, further distorting the market for rare literature.

As awareness of this practice grows, pressure may mount on Amazon to adopt less destructive methods of data acquisition. Alternatives, such as partnering with libraries or using existing digital archives, exist but may not offer the same volume of unique data. The incident serves as a cautionary tale about the unintended consequences of the AI arms race, where the pursuit of technological superiority can come at a significant cultural cost.

Sources behind this briefing

Go to the original reporting

  • Ars Technica↗Hidden Airtag reveals Amazon is trashing rare books to train AI
  • TechCrunch↗Amazon, which started off selling books, is destroying rare texts to train AI
  • Futurism↗Amazon Caught Destroying Rare Books to Train AI
  • the-decoder.com↗AirTag reveals how Amazon destroys rare books for AI training