ABSTRACT
How should law respond to transformative technologies whose benefits are inseparable from their risks? This Article argues that the answer lies not in binary choices between permission and prohibition, but in what might be called zones of contingent permission: legal frameworks that allow deployment of new technologies while requiring measures to detect and prevent specific harms. The concept emerges from an unlikely source: the ongoing litigation over whether and when training generative AI models on copyrighted works constitutes infringement.
Recent victories for Anthropic and Meta suggest AI training is fair use, but deeper analysis shows that will only sometimes be true. This Article explains why the boundary between permissible and infringing AI conduct will be jagged rather than smooth: the same computational process applied to a music model and a text model can produce dramatically different copyright risk, because memorization rates vary depending on a host of technical conditions and also because copyright’s threshold of substantial similarity that determines infringement is applied quite differently to some works compared to others.
That jagged frontier is only the beginning of the story. Drawing on Ronald Coase’s insight that legal rules are starting points for adaptation and negotiation rather than final allocations, this Article argues that the extensive literature on AI and copyright has focused almost exclusively on fair use while ignoring what comes next. AI companies that build systems to monitor and filter infringing outputs will not only reduce their legal exposure; they will create a deal point. The history of YouTube’s Content ID suggests that copyright safety infrastructure can become the foundation of a billion-dollar licensing industry benefiting developers, rights holders, and users alike.
Sag, Matthew, Copyright’s Jagged Frontier (February 28, 2026), Duke Law Journal (forthcoming).
Leave a Reply