i think this is a very large assumption.
What if in the future, AI models are as guarded as nuclear weapons? Because why doesn't this argument apply for nuclear weapons, but does for AI?