Musk has been making the point for a while now, and he repeated his opinion several times at a widely publicized long interview on stage at the the National Governors Association in Rhode Island this week.
He said he was exposed to the cutting edge of AI, and that it was the "scariest problem" facing human civilization. He it actually represented an "existential threat" to people, and he said it several times.
Musk said he didn't know of a solution for it.
He was also talking about electronic cars and solar power, and the discussion, complimented by several thoughtful questions from the governors in attendance, went into corporate and regulatory considerations in the 3 industries in question, but turned back to the dangers of AI frequently.
"AI is a fundamental risk to the existence of human civilization in a way that car accidents, airplane crashes faulty, drugs or bad food were not. They were harmful to certain individuals within society but not to society as whole,” he said.
Musk said AI so unusual that people should be pro-active in regulating it before it's too late. When he was asked by the governors for details about what sort of regulation, he seemed to retreat a bit from the position and called for more insight to be collected at the government level, because currently, it doesn't even have that.
He gave a few examples of the kinds of dangers AI posed, too, after saying the biggest threat was a deep level AI inside networks. One example that seemed to quiet the room was that an AI programmed to maximize stock value could invest in military, pull out of consumer goods, and shoot down a passenger plane over a warring region, while sending an anonymous tip-off that there was an enemy plane flying near the plane, a situation similar to what actually happened with the second Malaysian Air plane on the Ukraine-Russia border a couple years ago (although Musk said he did not believe that particular instance was necessarily any kind of conspiracy).