{"slug": "modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed", "title": "'Modern Poisoning Tools': AI Firms Buying and Destroying Thousands of Books To Feed AI Models", "summary": "AI companies are buying and destroying thousands of printed books to feed AI models, sparking copyright debates and cultural preservation worries. A federal judge ruled that Anthropic's scanning of legally acquired physical books is 'transformative' and eligible for fair use protection, while ISBNdb now facilitates orders of up to one million books for AI training, marketing pre-2022 print books as 'structurally clean of modern poisoning tools'.", "body_md": "# 'Modern Poisoning Tools': AI Firms Buying and Destroying Thousands of Books To Feed AI Models\n\n## The rise of AI models drives demand for printed books, sparking copyright debates and cultural preservation worries.\n\nArtificial intelligence companies are quietly fuelling a new market for printed books, but many of those volumes never reach readers. Instead, they are being dismantled, scanned and discarded, raising fresh concerns about copyright, cultural preservation and the growing appetite for high-quality data.\n\nWhat makes the practice especially striking is that many of the books being targeted are older, out-of-print or difficult to replace.\n\nFor booksellers, the sudden surge in demand has created an unexpected windfall. For critics, it reflects an industry increasingly willing to sacrifice physical works in pursuit of better AI models.\n\n## From Library Shelves to Industrial Scanners\n\nAs competition intensifies among developers of large language models, companies are searching for reliable training material that predates the explosion of AI-generated online content.\n\nPrinted books have become an attractive source because they are generally viewed as cleaner and more authentic than much of today's internet, which is increasingly saturated with AI-written text.\n\nWhile some AI developers have relied on digital libraries, others have turned to purchasing physical books in bulk.\n\nCourt records from a recently settled copyright case revealed that Anthropic bought large numbers of printed books from second-hand sellers before removing their bindings with a hydraulic cutting machine. The pages were then scanned using industrial imaging equipment so the contents could be converted into digital files for AI training.\n\nThe books themselves were destroyed during the process.\n\nThe legal argument centred on the US first-sale doctrine, which allows buyers to dispose of their purchased property as they choose. Because Anthropic digitised legally acquired physical books rather than redistributing copies.\n\nA federal judge found the scanning process to be 'transformative', making it eligible for fair use protection under US copyright law. That ruling has become one of the defining legal decisions shaping how AI companies acquire training material.\n\nThe company has also faced broader copyright disputes. In June 2025, [Anthropic agreed to a settlement valued at up to $1.5 billion in litigation involving authors](https://www.ibtimes.co.uk/anthropic-historic-settlement-pirated-books-1810392) over the unauthorised use of books, although the separate fair use ruling concerning scanned print books has continued to attract close attention.\n\n## Older Books Become AI's Most Valuable Resource\n\nThe growing demand has opened an unexpected business opportunity for companies that once served libraries and book distributors.\n\nISBNdb, which describes itself as operating the world's largest book database, has openly repositioned part of its business towards supplying AI developers with printed books. According to *404 Media*, the company now facilitates orders ranging from 1,000 to one million books at a time. Its sales pitch reflects how valuable pre-AI publications have become.\n\n'The world's best AI training data is setting on a shelf,' the company states on its website.\n\nISBNdb argues that books published before the widespread adoption of large language models offer a unique advantage because they have not been influenced by AI-generated writing.\n\n'Print books from the pre-LLM era are structurally guaranteed to be free of this contamination. That alone is a significant advantage,' the company explains. It added that physical books published before 2022 are 'structurally clean of modern poisoning tools', referring to concerns that AI-generated content circulating online could reduce the quality of future AI training datasets.\n\nThe company also appears aware that the process carries reputational risks.\n\n'The optics problem is real,' its website acknowledges. ''AI company destroys two million books' is not a headline that generates sympathy.' That unusually candid admission highlights an uncomfortable reality.\n\nEven businesses profiting from the trend recognise that public perception may prove as challenging as the legal questions surrounding it.\n\n## Rare Books Could Be Lost Forever\n\nBooksellers say the shift has become increasingly visible.\n\nOne independent dealer told *404 Media* that sales jumped dramatically in April, climbing from fewer than 20 books each week to several hundred almost overnight. The buyers selected books with little obvious connection beyond the fact that each carried an ISBN, leading the seller to suspect they were purchasing for AI laboratories.\n\nHis concern extends beyond the scale of the orders.\n\nMuch of his stock consists of rare, foreign-language and out-of-print titles, raising the possibility that some of the destroyed books may be among the few surviving copies still circulating through the commercial market.\n\n'I personally have mixed feelings about all of this,' the bookseller said. 'It benefits me financially as well as by clearing out old inventory that is otherwise unlikely to sell... On the other hand, I don't like the end-use, and I don't like that uncommon books are being pulped.'\n\nSimilar reports are now emerging outside the United States. According to *404 Media*, rare booksellers in the Netherlands have also experienced unexplained bulk purchases that many believe originate from AI companies, although no direct evidence has confirmed the identities of the buyers.\n\nThe secrecy surrounding these acquisitions only deepens the uncertainty. Companies such as ISBNdb promise anonymity for large purchasers, leaving booksellers to infer the destination of their stock from unusual buying patterns rather than confirmed identities.\n\nThe growing market illustrates how AI development is reshaping industries far beyond Silicon Valley. What was once a trade built around preserving and circulating knowledge now finds itself supplying raw material for machine learning.\n\nThe legal framework may currently permit the practice, but the destruction of increasingly scarce printed works raises questions that copyright law alone cannot answer.\n\n© Copyright IBTimes 2025. All rights reserved.", "url": "https://wpnews.pro/news/modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed", "canonical_source": "https://www.ibtimes.co.uk/ai-companies-demand-printed-books-copyright-concerns-1810831", "published_at": "2026-07-27 13:44:51+00:00", "updated_at": "2026-07-27 14:10:23.836664+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "ai-ethics", "large-language-models", "ai-infrastructure"], "entities": ["Anthropic", "ISBNdb", "404 Media"], "alternates": {"html": "https://wpnews.pro/news/modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed", "markdown": "https://wpnews.pro/news/modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed.md", "text": "https://wpnews.pro/news/modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed.txt", "jsonld": "https://wpnews.pro/news/modern-poisoning-tools-ai-firms-buying-and-destroying-thousands-of-books-to-feed.jsonld"}}