First, he fears what he does not understand. He can't even code.

Second, AI has already suggested 40,000 new possible chemical weapons in just six hours. So nuclear weapons are not necessary for AI to kill all humans.

https://www.theverge.com/2022/3/17/22983197/ai-new-possible-chemical-weapons-generative-models-vx

Third, AGI, an independend AI is decades away. Right now AI is just doing what we instruct it to do. That alone is concerning enough.