The problem is that you always need stronger AI to review weaker one, otherwise reviewed AI will eventually prompt-inject reviewing AI.
Alternatively they could also both escalate and go off the rails while warring with each other.
Such a model doesn't yet exist though, of course.