Anthropic has called three Chinese AI labs’ efforts to distill Claude “illicit.” The irony is that when authors sued Anthropic over their its use of their books to train Claude, Anthropic argued that training was fair use (and won on that issue). If it now argues that a competitor can’t learn from Claude’s outputs, it sounds like they're arguing against their own position. The cases aren’t identical, and fraudulent access changes the analysis. But in order to go after distillation, Anthropic will have to explain why the legal arguments that protect its own training shouldn’t protect someone else’s. @dbarabander ,@PossibltyResult and I joined @JacobRobinsonJD to talk through this tension, as well as what comes next in the fight over distillation.