" Again, wrong target. Humans need to *use the tools carefully*."
You are assuming humans are rational agents, and in addition act in-good-faith. You make the classic "Platonic" error, that virtue = knowledge. You imply < know the proper use of AI & all's well >. A sadly false assumption . More robust and lower-risk is to make the Calvinist error : that all humans are fundamentally irrational & corrupt ( in dealings with other people ). Then "physical" kill-switches make sense : kill-switches on AI ... kill-switches on insurance ... kill-switches on banking ... kill-switches on physicians : of-course the phrase [ kill-switch ] needs appropriate interpretation(s), but the meme is clear --- self-limiting tools ---> "this far and no farther".