Is an AI reimplementation a "clean room" implementation? What if the AI only generates pseudocode and a human implements the final code based on that? Etc etc ad infinitum.
Lawyers will be having fun with this philosophical question for a good decade.