ISBNdb Pulls AI Book-Training Page as $1.5 Billion Anthropic Case Fuels Rare-Book Backlash
Updated
Updated · Kotaku · Aug 1
ISBNdb Pulls AI Book-Training Page as $1.5 Billion Anthropic Case Fuels Rare-Book Backlash
3 articles · Updated · Kotaku · Aug 1
Summary
ISBNdb said it removed a marketing page that appeared to connect AI firms with printed books for training, insisting it has never trained models and never launched such a service.
The retreat followed outrage over reports that suppliers were getting unusual bulk orders for books that could be scanned and destroyed to create cleaner training data while avoiding copyright risks.
A deleted ISBNdb pitch had called books "the world's best AI training data," reinforcing suspicions after 404 Media reported on sellers unloading uncommon titles despite discomfort with the books being pulped.
The scrutiny widened after unsealed Anthropic "Project Panama" filings showed book scanning and destruction was treated as legally permissible; Anthropic later settled with authors for $1.5 billion.
The episode underscores a broader AI-data squeeze as companies hunt higher-quality, non-web sources and may favor destructive scanning simply because it is faster and cheaper.
Did a major metadata provider actually build a secret pipeline to feed physical books to AI, or was it truly just a market test?
Why are mysterious buyers bulk-ordering and destroying thousands of physical books, and what does this mean for the future of human-written literature?