The framework would divide the world into AI “haves” and “have nots,” based on nations’ willingness to align with the U.S.