AI Companies Destroying Millions of Rare Books for Training Data
Key Takeaways
- ▸Anthropic's 'Project Panama' uses industrial-scale destructive scanning—hydraulic cutting machines and high-speed scanners—to digitize and destroy millions of rare books
- ▸ISBNdb coordinates bulk book purchases specifically targeting rare, out-of-print, and pre-2022 titles valued as 'contamination-free' AI training data
- ▸Destruction of rare books is irreversible; digitized versions become subject to AI interpretation rather than preserving original content in its authentic form
Summary
Major AI companies, particularly Anthropic, are engaging in large-scale "destructive scanning" operations where millions of rare and out-of-print physical books are purchased, digitized, and then destroyed. ISBNdb, a facilitator of these operations, explicitly targets older and specialist volumes published before 2022, valuing them as "contamination-free" training data as AI models increasingly feed on themselves. Internal documents revealed in Anthropic's copyright lawsuit exposed "Project Panama," an industrial operation using hydraulic cutting machines to slice book spines and high-speed scanners before shredding, with executives describing their goal as "disruptively scan[ning] all the world's books."
The practice threatens irreplaceable cultural artifacts. Small booksellers across Europe and the United States report sudden surges in bulk orders for rare and foreign-language titles—simultaneously benefiting sales while enabling destruction of books that have survived centuries. Once a book's final physical copy is shredded, it is lost forever, particularly catastrophic for rare pre-digital-era publications that may exist nowhere else. While a federal judge's ruling in the Anthropic copyright case determined destructive scanning qualifies as "transformative" fair use, critics argue that replacing physical originals with AI-processed digital versions fundamentally compromises both cultural preservation and scholarly integrity.
- A federal court ruling classified the practice as 'transformative fair use,' but the operation lacks transparency—ISBNdb offers NDAs and advises clients to frame destruction as 'digital preservation'
Editorial Opinion
The systematic destruction of irreplaceable rare books to fuel AI model training represents a troubling collision between technological acceleration and cultural stewardship. While companies frame this as 'digital preservation,' they are effectively replacing irreplaceable originals with AI-processed interpretations—a trade-off that leaves no room for future discovery, scholarly reexamination, or preserving human knowledge in its most authentic form. This practice demands urgent legislative action to protect cultural heritage before millennia-old knowledge is permanently lost.
