I think there's quite a bit of "reverse alignment" going on there, essentially the humans will generally not even ask the AI to do something they'd be unwilling

I think there's quite a bit of "reverse alignment" going on there, essentially the humans will generally not even ask the AI to do something they'd be unwilling to do, partially accomplished through the control of language and thought.