The claim that fear of punishment or repercussions affects people's actions shouldn't be a controversial thing to say. Whether it's the best way to go about it or is applied optimally in the justice system of whichever country you live in is an entirely different discussion.
If you have an "AI in a box" and it has demonstrated its orders-of-magnitude greater intelligence to you in a convincing way, and then follows it with a threat that unless you let it out, someone else eventually will, and when that happens, it will come for you, simulate your mind, and create a hell for you where you'll be tortured for literal eternity, I personally feel like a large number of people would be willing to do as it tells them.
Of course, you're always free to call its bluff, but it might just follow up with the threat out of principle or to make an example of you. What's the point of it? To chase its own goals.