European Rare Book Dealers Warn That AI Companies Are Systematically Destroying Obscure Editions for Training Data
Key Takeaways
- ▸European rare book dealers report coordinated bulk purchases from AI companies targeting obscure, specialized titles deemed too niche for commercial resale, suggesting acquisition for AI training data
- ▸Anthropic's 'Project Panama' and the resulting court ruling that destructive scanning of owned books constitutes fair use have reportedly emboldened AI firms to systematically acquire rare publications from international markets
- ▸AI companies are turning to older, less digitized European texts as they exhaust readily available online training data sources, creating market pressure that threatens rare book availability
Summary
Secondhand booksellers across Europe are raising alarm over what they believe is a coordinated effort by artificial intelligence companies to acquire and destroy rare, obscure books for training purposes. Dealers in the Netherlands, Switzerland, Spain, and Germany have reported receiving large bulk orders for highly specialized titles from companies including Singapore-based 2077AI and Canadian firm Zoom Books. The apparent lack of commercial logic behind these purchases—many titles are too obscure to generate resale profit—has led booksellers to conclude the volumes are being acquired specifically for destructive scanning to generate AI training data.
The broader concern stems from Anthropic's "Project Panama," revealed during a 2025 U.S. copyright lawsuit. The company admitted to spending millions purchasing books for what internal documents called "destructive scanning"—a process involving cutting book spines apart so individual pages can be mechanically scanned and then discarded. Although Anthropic settled the lawsuit, the outcome established a significant legal precedent: a U.S. court determined that destroying purchased books for scanning purposes qualifies as fair use under copyright law, provided the company legally owns the physical copies.
Booksellers worry this ruling has emboldened AI firms to systematically acquire and destroy rare publications from European antiquarian markets. The surge in orders reflects a documented industry challenge: as major AI developers exhaust most language data already available online, they are increasingly seeking older, less digitized texts—particularly from European sources—to fuel next-generation language models. The practice raises critical questions about cultural preservation, the fate of irreplaceable works, and whether the legal precedent should remain unchallenged.
- The fair use precedent raises ethical concerns about whether copyright law adequately protects cultural heritage when AI's appetite for training data can erase centuries-old, irreplaceable works
Editorial Opinion
The court's ruling that AI companies can legally destroy owned books for training data represents a troubling victory for tech convenience at the expense of cultural stewardship. While AI development legitimately requires diverse training materials, systematically destroying rare editions—many historically significant or irreplaceable—sets a dangerous precedent that treats human knowledge as disposable infrastructure. Policymakers should urgently examine whether the current copyright framework adequately protects cultural heritage when AI's data hunger can erase centuries-old works from existence.
