Published: July 29, 2026
By: TAD Editorial Team
⸻
NEW YORK —
As artificial intelligence developers race to build more capable AI systems, a growing number of technology companies and data brokers are reportedly purchasing massive quantities of physical books—not to preserve them, but to dismantle them.
Industry investigations suggest that warehouses filled with second-hand, rare and even out-of-print books are being converted into training material for AI models. Instead of carefully archiving the printed volumes, many are having their bindings cut away so individual pages can be rapidly scanned before the remaining paper is discarded or shredded.
The practice has sparked criticism from authors, historians and preservation advocates, who argue that while the books may be legally purchased, the destruction of physical copies represents an irreversible cultural loss.
⸻
A New Hunt for ‘Human-Written’ Knowledge
The latest wave of large language models requires enormous amounts of written material to improve reasoning, language understanding and factual accuracy.
However, researchers say the internet has changed dramatically since generative AI became mainstream.
Today, websites, blogs and online forums are increasingly filled with AI-generated articles, rewritten content and automated text—often referred to by critics as “AI slop.”
Because newer AI systems risk learning from machine-generated information instead of original human writing, developers are reportedly searching for datasets created before AI content became widespread.
Books published years before the current AI boom are therefore considered some of the most valuable remaining sources of authentic human-written material.
Investigative reporting indicates that technology companies are now relying on intermediaries and specialized data brokers to acquire enormous collections of printed books for digitization.
⸻
Why Companies Are Cutting Books Apart
Unlike libraries that carefully preserve books during digitization, many commercial scanning operations reportedly use what is known as destructive scanning.
The process begins by removing a book’s spine so every page can pass through industrial document scanners at extremely high speeds.
While non-destructive scanning methods keep books intact, they require manual page turning and significantly more labor, making them slower and more expensive for large-scale projects.
By contrast, destructive scanning allows thousands of pages to be processed in a fraction of the time.
Once the pages have been digitized, the remaining paper is often recycled, shredded or discarded.
The resulting digital text is then stored inside proprietary databases used to train commercial AI models rather than public digital archives.
⸻
Massive Demand Behind the Scenes
Reports published by technology media outlets indicate that demand for printed books has expanded dramatically over the past year.
Court documents and investigative reporting have suggested that intermediaries involved in supplying AI training data have been negotiating purchases involving hundreds of thousands—and in some cases millions—of physical books.
Rather than acquiring books individually, buyers reportedly source large collections from wholesalers, liquidation sales, libraries, estate collections and used-book distributors.
Rare titles, academic works and books no longer available through traditional publishers are said to be particularly valuable because they contain writing styles and information not widely available elsewhere online.
⸻
Legal Books, Ethical Debate
The practice appears to operate within existing property laws because the physical books are generally purchased through legal commercial transactions.
Owning a printed book gives buyers the right to destroy that physical copy.
However, the broader debate centers not on ownership, but on preservation.
Historians argue that rare books represent more than collections of printed words—they are physical cultural artifacts with historical value.
Authors and publishers have also questioned whether digitizing books for AI training should require additional permissions beyond purchasing the physical copy.
Meanwhile, several ongoing copyright lawsuits involving AI companies continue to test how existing intellectual property laws apply to machine learning datasets.
Those legal battles could help shape how future AI developers obtain training material.
⸻
The Race for Cleaner Data
As AI-generated content spreads across the internet, many researchers believe access to reliable human-created writing will become increasingly valuable.
Training future AI models on content already produced by earlier AI systems raises concerns about declining quality, factual errors and repetitive language patterns.
This has encouraged developers to seek older, carefully edited sources—including books, newspapers and academic publications—that predate today’s AI-driven internet.
For technology companies competing to build increasingly advanced AI systems, access to cleaner datasets may prove to be one of the industry’s most valuable competitive advantages.
⸻
Why It Matters
The reported large-scale destruction of printed books illustrates a growing tension between technological progress and cultural preservation. While AI companies require enormous volumes of high-quality text to improve future models, historians and preservation advocates warn that physical books—especially rare editions—carry historical and educational value beyond the information they contain digitally.
⸻
TAD Perspective
Artificial intelligence is reshaping how knowledge is collected and processed, but the methods used to build tomorrow’s AI systems are drawing increasing public scrutiny. The debate is no longer limited to copyright; it now includes questions about preserving cultural heritage, balancing technological innovation with historical responsibility, and determining whether society should sacrifice physical archives in pursuit of digital progress.
Sources
Tom’s Hardware
404 Media (as referenced in reporting)
Techworm
Industry reports and court filings
