Background & Context§
The insatiable demand for high-quality training data has driven AI companies to increasingly extreme measures. A controversial service called ISBNdb facilitates anonymous bulk orders of up to one million books, enabling clients to acquire physical copies for high-speed scanning—a process that destroys the original books. According to Hedgie (@HedgieMarkets), these practices specifically target rare books, including those with fewer than a dozen surviving copies. A federal judge recently ruled that this destruction constitutes fair use because eliminating the original ensures only one digital copy exists at a time. This ruling has effectively created a new business model centered on the systematic destruction of cultural artifacts for AI training.
The News: What Happened Exactly§
Hedgie's report on July 27, 2026, details that AI companies are bulk-buying rare books and feeding them through high-speed scanning machines that cut off the spines and shred the originals. ISBNdb, a service that facilitates orders of up to a million books, keeps buyers anonymous and even offers NDAs as a feature. The company's website explicitly acknowledges the potential public relations disaster, stating that "'AI company destroys two million books' is not a headline that generates sympathy." Despite this awareness, they have built an entire business around enabling these transactions quietly, coaching clients to call the process "digital preservation."
Anthropic has hired the former head of Google Books partnerships specifically to obtain "all the books in the world," signaling the scale of these ambitions. The preference is for books published before 2022, as they are guaranteed free of AI-generated text, making them premium training data. A bookseller told 404 Media that rare books with almost no surviving copies are being fed into this pipeline—books that survived wars, fires, and centuries of handling are being shredded so an AI can learn to write better marketing emails.
The legal underpinning comes from a federal judge's ruling that destroying the original is fair use because it ensures only one copy exists at a time. This reasoning is controversial: while copyright law traditionally focuses on unauthorized copying, the physical destruction of rare artifacts sidesteps typical fair use considerations. The ruling effectively incentivizes destruction by removing the possibility of multiple copies existing, which could otherwise lead to copyright infringement claims. Critics argue that this misinterpretation of fair use could accelerate the loss of irreplaceable cultural heritage.
Historical Parallels & Similar Incidents§
This practice echoes the infamous "Google Books" scanning project from the mid-2000s, which also faced legal challenges over copyright. However, key differences distinguish the current situation. Google partnered with major libraries to scan public-domain and copyrighted works, but explicitly did not destroy originals. The Authors Guild lawsuit against Google in 2005 resulted in a landmark fair use decision in 2016, allowing Google to display snippets of copyrighted books. Crucially, Google preserved the physical copies and even returned them to libraries after scanning. The current wave of destruction inverts this: rather than preserving cultural artifacts, companies destroy them to prevent legal exposure.
Another parallel is the systematic theft of rare books for profit, such as the case of Marino Massimo De Caro, who stole thousands of rare books from Italian libraries between 2011 and 2012, selling them or destroying them to cover his tracks. While De Caro's motivations were monetary, the destruction in AI training is arguably more systemic and legally sanctioned. The difference is that De Caro operated covertly; today's companies operate with explicit judicial approval, making the practice scalable and potentially leading to the extinction of entire rare book collections.
The historical lesson is clear: when rare artifacts are destroyed en masse, the loss is permanent. The burning of the Library of Alexandria is often invoked as a cautionary tale, but that was a single catastrophic event. The current trend is a slow, systematic destruction fueled by corporate demand for training data. As noted by user @MoonlitMaven, this is "Burning of the library of Alexandria 2.0." Unlike the accidental fire in Alexandria, this destruction is deliberate and ongoing, with a perverse legal justification.
# A simple model of lost rare book count
rare_books_lost = 0
for year in range(2024, 2030):
# Estimated 100,000 rare books shredded per year by AI companies
rare_books_lost += 100000
print(f"Year {year}: {rare_books_lost} rare books lost irreversibly")The above code represents a conservative estimate; the actual number could be higher as ISBNdb enables orders of up to a million books. The AI industry's data hunger, combined with a legal framework that rewards destruction, sets a dangerous precedent. As @PG_Ting noted, "The physical copies are artifacts that need to be preserved, not destroyed. We are becoming increasingly stupid, careless and thoughtless, everywhere you look."
In response, Elon Musk announced that the SpaceXAI team will "preserve any rare books in a library and scan them the hard way vs just cutting off the spine and scanning," offering a rare corporate pushback. However, unless the legal ruling is overturned or industry standards change, the destruction is likely to accelerate. The trade-off between AI advancement and cultural preservation has never been starker, and the current trajectory suggests that irreplaceable knowledge will continue to be sacrificed for marginal improvements in language model performance.