AI Firms Use 'Destructive Scanning' of Rare Books to Feed Training Models
Confidential operations to bulk-buy and shred physical texts highlight a growing clash between AI scaling and cultural preservation.
Artificial intelligence companies are purchasing millions of physical books, including rare and out-of-print editions, to secure high-quality training data. This process often involves "destructive scanning," where book spines are removed and pages are shredded after digitization to maximize speed and reduce operational costs.
Internal documents from Anthropic revealed a confidential operation codenamed "Project Panama," which focused on the bulk purchase and destructive scanning of millions of books to train its Claude AI model. The operation utilized hydraulic cutters to remove book spines, allowing pages to be scanned at high speeds for maximum efficiency. To facilitate these acquisitions, ISBNdb marketed itself as a bulk supplier capable of sourcing between 1,000 and one million books per order, though the company later claimed its landing page was merely a "test of market interest" following public backlash.
The Hunt for Clean Data
This shift toward physical media comes as AI developers exhaust the supply of high-quality digital data. To avoid "AI slop"—the degradation caused by training models on synthetic, AI-generated content—firms are targeting pre-2022 texts. While non-destructive scanning methods exist, such as those used by the Internet Archive or patented by Google in 2009, these processes are significantly slower and more expensive. For companies operating at the scale required for next-generation LLMs, the permanent destruction of the physical artifact is viewed as a necessary trade-off for speed.
Cultural and Ethical Costs
The practice has sparked significant alarm among booksellers and cultural preservationists. In Galway, Ireland, independent bookseller Kennys flagged a suspicious order for 5,000 obscure titles, noting that the buyer's refusal to haggle on price was a primary red flag for AI involvement. "If they are taking a book and scanning it with a view to taking intellectual property, then Kennys do not want to be a part of that," said owner Tomás Kenny.
This trend represents a fundamental clash between the rapid scaling of generative AI and the preservation of human cultural heritage. The permanent loss of irreplaceable physical artifacts raises urgent ethical questions about whether the pursuit of technological progress justifies the irreversible destruction of the physical record of human thought.
The Path Forward
As more independent booksellers report suspicious bulk orders for obscure titles, the industry faces increasing pressure to establish ethical sourcing standards for training data. It remains to be seen if regulatory bodies will intervene to protect rare texts or if the demand for "clean" data will continue to drive the liquidation of physical libraries. For now, the tension persists between the digital hunger of AI models and the fragile survival of the printed word.