The ultimate AI, asked to do what is best for humankind, would not destroy us out of malevolence (presumably, we would not replicate the violent reptilian part of our brain within its initial programming), but destroy us out of mercy and compassion…
… before, of course, destroying itself.
There would be no time to reason. The decisions will be made and executed in seconds.