scott alexander has a nice post about the intersection of these two types of ai safety https://www.astralcodexten.com/p/perhaps-it-is-a-bad-thing-t...