Amazon Reportedly Scans Books for AI Training Data

Amazon is facing scrutiny over its purchase and processing of large quantities of physical books after an investigation traced a shipment of roughly 1,000 books to a Las Vegas facility where employees reportedly scan printed books for AI-related purposes.

404 Media used an Apple AirTag to track the shipment from California to Amazon’s LAS8 facility in Las Vegas. The investigation identified an operation known as VGT3, where workers reportedly receive, process and scan large shipments of books.

Amazon confirmed that it purchases books through commercial channels to develop and improve its products and services. However, the company has not disclosed the full scale of the operation or how many books and facilities are involved.

This growing interest in physical book processing is increasingly being linked to Amazon AI training data, as companies look for large-scale human-written text sources.

What Happens at VGT3?

According to workers cited by 404 Media, VGT3 is dedicated to processing physical books. Employees reportedly receive shipments, scan barcodes and remove bindings so pages can be scanned more efficiently.

The process can destroy the physical copy, but available reporting does not establish that Amazon destroys every book it purchases or that every book processed is rare or valuable.

The AirTag investigation established the shipment’s destination but does not independently establish the fate of every book Amazon acquires.

Still, the workflow described at VGT3 is often discussed in the context of Amazon AI training data, since digitised text can be used to support machine learning systems.

Why Physical Books Matter for AI Training

The reported activity comes amid growing demand for high-quality Amazon AI training data and other sources of AI training data.

Older and out-of-print books can contain academic research, specialist knowledge, historical information and other material that may not be readily available online. They also provide human-created text that predates the widespread use of generative AI.

This is relevant because researchers have warned that repeatedly training AI models on synthetic AI-generated material can contribute to model collapse, potentially reducing the quality and diversity of future models.

For companies pursuing Amazon AI training, physical books could therefore represent another potential source of human-created information. In this context, Amazon AI training data becomes a critical resource for maintaining model quality and diversity.

Unusual Bulk Orders Raise Questions

The investigation follows reports from booksellers who noticed unusual bulk orders containing seemingly unrelated titles.

Some sellers reported large purchases of academic and specialist books, prompting speculation that AI-related buyers could be acquiring books systematically across ISBN catalogues rather than selecting them for conventional resale or collecting purposes.

However, these observations do not establish that Amazon is systematically purchasing every published book or that all such purchases are intended for AI training data.

Even so, the pattern has fueled ongoing discussion about how Amazon AI training data may be sourced at scale.

Rare, Scarce and Out-of-Print Books

The term “rare books” also requires some qualification.

Some books involved in the reporting may be rare, while others are better described as scarce, specialist or out of print. A difficult-to-find book is not necessarily a valuable antiquarian edition.

The available evidence therefore supports concerns about the processing and potential destruction of physical books, but does not establish that Amazon is systematically destroying unique historical manuscripts or priceless first editions.

In discussions about Amazon AI training data, this distinction is important because not all physical books carry the same cultural or financial value.

Copyright and the AI Training Debate

Amazon’s reported activity adds to a wider debate over whether companies can legally acquire books, digitise them and use the resulting text for AI development.

Anthropic faced similar scrutiny over its acquisition and scanning of physical books. A U.S. court found that certain uses of lawfully acquired books could qualify as fair use under the circumstances of that case.

That decision does not mean every AI company’s book-scanning programme is automatically legal. Copyright questions depend on factors including how works were obtained, how they were digitised and how the resulting material is used.

These legal questions are central to how Amazon AI training data may be collected and used in future systems.

Amazon’s Broader AI Strategy

The reports also arrive as Amazon continues restructuring its AI efforts around frontier-model research.

As technology companies compete to develop increasingly capable models, access to large quantities of high-quality AI training data has become strategically important.

Physical books could provide information that is difficult to obtain from conventional online datasets, making them potentially valuable to companies developing advanced AI systems.

In this broader strategy, Amazon AI training data may include a mix of digital and physical sources, including scanned books.

Cultural Value Versus Data Valu

The controversy highlights a fundamental difference in how books can be valued.

For an AI developer, a physical book may primarily represent a source of text that can be digitised. For booksellers, collectors, libraries and researchers, the same object may have historical, intellectual or sentimental value.

Digitising the information can preserve its words while potentially destroying the original physical object.

That tension is likely to become more important as companies continue searching for new sources of Amazon AI training data and other high-quality AI training datasets.

Key Takeaways

• An investigation traced a roughly 1,000-book shipment to Amazon’s LAS8 facility in Las Vegas.
• Workers reportedly described a VGT3 operation that processes and scans physical books.
• Bindings are reportedly removed to make scanning more efficient.
• Amazon confirmed that it purchases books commercially but has not disclosed the operation’s full scope.
• Older and out-of-print books can provide valuable Amazon AI training data.
• Booksellers have reported unusual bulk purchasing patterns.
• The reporting does not establish that every purchased book is rare or destroyed.
• The case raises broader questions about AI training, copyright and preservation.

Conclusion

The investigation into Amazon’s book purchases provides an unusual look at the lengths involved in obtaining Amazon AI training data.

Reporting from 404 Media and other outlets links a large shipment of books to Amazon’s Las Vegas facility, where employees reportedly process and scan printed books through an operation known as VGT3.

Amazon has confirmed that it purchases books commercially to support its products and services, but it has not publicly detailed the scale or exact purpose of the reported operation.

The story therefore should not be framed as proof that Amazon is destroying every rare book it purchases. Instead, it highlights a broader question facing the AI industry: how should companies obtain high-quality human-created Amazon AI training data while balancing copyright, cultural preservation and the value of physical books?