Full Breakdown
AI Companies’ Destructive Scanning of Rare Books Sparks Global Outcry
8/21/2026, 5:23:23 AM
Core Event: Destructive Scanning of Physical Books for AI Training
An investigation by 404 Media revealed that a shipment of rare books sent to an Amazon fulfillment center in Las Vegas was not packaged for resale. Amazon staff removed the bindings and cut the volumes to create digital scans. The report says the same procedure has been applied to “numerous other books,” prompting criticism from cultural-heritage groups worldwide.
Background & Context: Data Hunger and the Rise of “Project Panama”
Large language models need massive, high-quality text corpora. As publicly available internet content became saturated with AI-generated material, firms turned to out-of-print and rare titles. Anthropic, creator of the Claude series, labels this effort “Project Panama” in internal documents, noting the codename is meant to conceal the activity.
Data & Statistics: Scale of Acquisitions and Financial Stakes
- Amazon’s physical-book processing – The investigation documents at least one shipment of rare books that were cut for scanning; the exact number of titles affected is undisclosed.
- Anthropic’s internal target – The April 2024 memo describes a universal scanning mandate, implying a scope that could encompass millions of volumes.
- Industry-wide bidding – The Spirit-data deal outbid data-labelling startup Mercor, which had offered $7.5 million, illustrating the premium placed on proprietary corporate archives.
Official Statements & Responses
- Anthropic – A spokesperson, quoted by Fortune, said the company is not buying “rare or antiquarian books” for scanning and destroying them. The internal memo, however, describes a universal scanning program.
- Google – A representative said the Spirit Airlines data will be anonymized before incorporation into Google’s AI models, emphasizing compliance with privacy standards.
- Amazon – The company has not issued a public comment on the 404 Media findings.
Conflicting Reports & Gaps
- Destruction vs. Scanning – Some outlets reported that Anthropic shreds books after scanning; the Fortune spokesperson denied shredding, while the memo indicates books are cut as part of scanning, leaving final disposition unclear.
- Scope of Amazon’s activity – The investigation confirms destructive scanning of a specific shipment but does not quantify additional titles.
- Transparency of data-use policies – Neither Anthropic nor Google has provided detailed methodology for how scanned text will be filtered, de-duplicated, or protected against inadvertent exposure of copyrighted material.
Why It Matters: Implications for Cultural Heritage and AI Regulation
Physically dismantling rare books to create training data raises concerns about preserving cultural artifacts and the ethical sourcing of AI inputs. Large-scale scanning without oversight could lead libraries and collectors to lose irreplaceable works. The willingness of major tech firms to spend millions on proprietary data also signals a shift toward closed-source training material, potentially widening the gap between well-funded AI developers and the broader research community.
What’s Next: Potential Oversight and Industry Response
Industry observers note the controversy may prompt legislative or regulatory scrutiny of how AI developers acquire and process copyrighted or heritage material. No formal investigations have been announced, but the public backlash and high-value data deals suggest policymakers could consider new guidelines for the acquisition and handling of source material used in AI training.
