AI Firms Suspected of Buying and Destroying Rare Books for Training Data

Secondhand and rare booksellers across the UK and Ireland are sounding the alarm over a surge in suspicious bulk orders of obscure titles. Industry professionals believe AI companies are aggressively acquiring physical books to scan for high-quality training data, often using destructive methods that permanently erase the originals.

The trend has manifested in procurement patterns that deviate from traditional collecting. Tomás Kenny of the Irish bookstore Kennys flagged a particularly suspicious order for 5,000 obscure titles. According to Kenny, the buyer refused to haggle on price—a behavior he identifies as a red flag for AI-driven procurement rather than human collecting. Kenny described the impact of AI on the trade as "terrifying our industry," stating that if firms are scanning books to seize intellectual property, his shop does not want to be a part of that process.

The Race for Clean Data

This shift toward physical media comes as AI developers exhaust the available high-quality digital data on the open web. To improve Large Language Models (LLMs), firms are seeking long-form, structured text that is free from "AI slop"—the recursive, low-quality content generated by other AI models that now plagues the internet. Physical books represent one of the last remaining reservoirs of verified, human-authored prose.

While some industry leaders claim to prioritize preservation, other evidence suggests a more aggressive approach. A lawsuit, Bartz v. Anthropic, revealed a secretive operation codenamed "Project Panama." The filings allege that Anthropic scanned and destroyed millions of physical books to train its models, utilizing destructive scanning techniques—such as cutting off spines—to maximize processing speed and reduce costs. Further complicating the landscape, the book-database company ISBNdb has reportedly advertised services specifically designed to help AI firms source books in bulk for training purposes.

Cultural and Legal Stakes

The practice creates a profound moral and economic dilemma for the secondhand book trade. Sellers must weigh the immediate financial gain of massive bulk sales against the permanent loss of rare physical editions and the erosion of authors' rights. Beyond the financial impact, the destruction of these volumes represents a permanent loss of literary history and physical intellectual property.

What Remains Unconfirmed

As the industry reacts, the full scale of "Project Panama" and similar operations remains unclear. While the Bartz v. Anthropic lawsuit provides a glimpse into the practice, many AI firms continue to operate their data acquisition strategies in secret. The industry is now watching to see if regulators will classify the destructive scanning of copyrighted physical works as a distinct form of intellectual property theft or if the practice will continue unchecked in the race for model dominance.