I'm not really sure what people see in the idea of hyperpersuasiveness. The relatively small intelligence differential between humans and other monkeys has led to such a difference in outcomes that we don't really need to try to persuade them. If the gap between meat's ability to handle statistics and silicon's ability to handle statistics turns out to be a similar power gap there won't be much persuading needed. Look at the gap European civilisation managed to open up over everyone else militarily through just organisational techniques.

The likely path to an AI going rogue is big country vs small country tension (say, China-Taiwan type scales). Small country gets desperate, follows some path to build autonomous, evolving AI for defence and deterrence. I can't think of a precise scenario where that would make sense, but it'll happen somewhere.

AI doesn't need to persuade its way out of the box, we just need some desperate group somewhere who's choices are annihilation or rolling the dice that an AI is a gentler overlord. There are lots of places where that is the case - the same calculus comes in to play with nukes and we see many countries nuking up over the last century - despite the fact that we're putting "end of civilisation" on the table as an option because of it.