Source: MIT Sloan

U.S. law remains unresolved on whether training AI models on copyrighted works constitutes fair use or infringement, and the U.S. Copyright Office has not issued a binding determination. Thomas Malone, Frank Pasquale, and Andrew Ting propose “learnright” laws that would create an exclusive right for copyright holders to license their works specifically for AI model training, arguing this would enable creators to claim a share of revenues from generative AI systems. The authors outline an opt-in registration program whereby creators register copyrighted content and AI companies obtain licenses, negotiating compensation either directly with creators or via their literary or artistic agents to reduce transaction complexity.

The authors cite ongoing litigation such as Encyclopedia Britannica’s lawsuit against OpenAI and previous cases involving Anthropic and Meta to illustrate legal uncertainty around fair use for AI; courts have applied flexible, contextual fair-use analyses but no generative AI case has reached the Supreme Court. They present three main rationales for learnrights: preserving creators’ incentives to produce original work, ethical acknowledgment and compensation for influence, and avoiding erosion of future content supply; they also argue that automatic, blanket learnright coverage would be impractical and that a robust market with reasonable licensing fees could emerge to benefit both creators and AI providers.