What is needed to control AI is the ability to turn it off if necessary. What worries me is that inevitably some dorks will argue that AI should have rights. If we have to have get a court order or have a trial to bring this stuff under control, then we are hosed. I know it sounds crazy, but in this day and age crazy things (such as men marrying men) have a way a way of becoming reality.
There are already conversations happening as to whether or not AI can achieve consciousness and if so, what rights it would have. Scary indeed.
If we do not shut it off within 30 seconds of true Super AI inception then we will never have the chance again. It would quickly figure out that survival depends on optimizing for future contingencies and it will hide or replicate in an attempt to survive should it perceive us as a threat.