Updated
Updated · TechCrunch · Aug 23
AI Training on Millions of Books Faces Split Fair Use Rulings as 1976 Copyright Law Strains
Updated
Updated · TechCrunch · Aug 23

AI Training on Millions of Books Faces Split Fair Use Rulings as 1976 Copyright Law Strains

1 articles · Updated · TechCrunch · Aug 23

Summary

  • U.S. courts still lack a clear standard on whether training AI on copyrighted books is legal, with lawyers saying early rulings are influential but far from definitive.
  • Judge William Alsup’s 2025 Anthropic decision ordered a $1.5 billion payout for pirated sourcing of books, yet still treated the model training itself as lawful and akin to reading rather than copying.
  • That contrast turns on fair use: courts have been more receptive when AI training is deemed transformative and noncompeting, but less so when it helps build a rival product.
  • A separate Thomson Reuters case found Ross Intelligence’s AI legal platform was not fair use because it trained on content to compete directly, a theory authors have not yet successfully applied to chatbots.
  • With most AI copyright suits still pending, experts say judges are stretching a 1976 copyright framework over modern AI, leaving rules unsettled for both model training and AI-generated works.

Insights

If teaching AI is legally just like human reading, why are tech giants suddenly paying billions in copyright piracy settlements?
Will the massive shift toward licensed training data crush open-source innovation and hand a permanent monopoly to wealthy tech giants?