Judge Rules AI Destructive Scanning of Books is Fair Use

Artificial intelligence developers are purchasing millions of vintage physical books, specifically those printed before 2022, to avoid synthetic AI-generated text and potential model collapse. These companies utilize a destructive scanning process, using industrial equipment to slice off book bindings and shred the remains after digitization to ensure high-speed data acquisition.

Anthropic launched an internal initiative called Project Panama, spending tens of millions of dollars to contract Datamation for these services to train its Claude chatbot. The practice gained legal momentum after U.S. District Judge William Alsup ruled on July 21, 2026, in Anthropic v. Authors Guild that digitizing legally acquired physical texts for model training is transformative and constitutes fair use under the Copyright Act.

To facilitate these acquisitions, companies use intermediaries like ISBNdb to make anonymous bulk purchases of thousands to millions of volumes. This trend has sparked outcry from librarians, booksellers, and preservationists, who warn that rare, foreign-language, and out-of-print editions are being permanently erased. While Google, Microsoft, and Harvard University have released one million public-domain digitized books as a non-destructive alternative, AI firms continue to employ the buy-scan-destroy pipeline to secure high-quality human-written content.