Not counting tests, we haven't seen one in action in over 80 years. If we could practice this level of caution with AI, that would be a great start.
Now of course we shouldn't completely deregulate nuclear power either. As in all things, the middle way.