Social media posts circulating online claim that artificial intelligence development companies are purchasing millions of books—including rare and hard-to-find editions—only to scan and systematically destroy them for AI model training. A detailed fact-check indicates that while key elements of the claim are true regarding specific company practices, broader assertions involving all AI firms targeting rare books remain unsubstantiated.
According to investigation reports and fact-checking organization Snopes, court documents unsealed during a copyright lawsuit revealed that US-based AI company Anthropic acquired large quantities of physical books with plans to scan and discard them.
The internal April 2024 documents refer to the initiative as “Project Panama.” To accelerate the digitization process, physical books were unbound by cutting away their spines, allowing individual pages to pass through high-speed automated scanners.
Key Findings of the Investigation:
- Project Panama Execution: Internal records from Anthropic confirmed the acquisition of millions of physical books through late 2024 to build proprietary training datasets.
- Destructive Scanning Method: High-speed scanning techniques required severing book spines, rendering the physical copies permanently destroyed post-digitization.
- The “Rare Books” Distinction: Documents showed a strategic preference for “less common” titles to diversify training data. However, it remains unverified whether “less common” strictly referred to out-of-print titles or simply less popular commercial publications.
- Industry-Wide Application: While evidence explicitly confirms Anthropic’s practices under court disclosures, independent verification does not exist to prove that all major AI developers are executing similar physical destruction programs on a mass scale.
📊 Fact-Sheet Summary
- Subject: Claims of AI companies buying, scanning, and destroying physical/rare books.
- Key Entity Named: Anthropic (“Project Panama”).
- Verification Status: Mostly True (Confirmed for specific entity via court records; unverified as a universal industry-wide practice).
- Scanning Technique: Destructive high-speed scanning (spine removal).
- Primary Source of Disclosure: Court-ordered filings in an ongoing copyright lawsuit (reported via Snopes).
