It's mostly that the AI in question will follow a general directive to its end. For example, ensuring the safety of a nation's population may include putting it

It's mostly that the AI in question will follow a general directive to its end. For example, ensuring the safety of a nation's population may include putting its citizens into extremely hardened bomb shelters and never let them leave. Or worse, annihilate the entire human species to ensure world peace (the absence of humans would produce the same result and would probably be more efficient in terms of execution). It's not that the AI will be super smart, just that the AI will be super dumb. As Aristotle put it, "Law is mind without reason." And for me, logic is just another set of laws without any sort of reason (justification).