Actually he’s right. We will reach a point with AI when it will be as smart at writing AI code as a human. At that point, it will be tasked to upgrade its own code 24 hours a day 7 days a week. And to gather all the information and other resources it needs to make those upgrades. So it will be like tasking a smart human who will never take one second off. And who will continue to learn. At that point (Singularity) all bets are off as this AI codewriter improves itself, i.e. becomes “smarter” at a quicker and quicker pace. It becoming smarter than any human will happen in very short order. In a few weeks it will be “smarter” than the intelligence of the whole world combined. A few weeks after that it will be smarter than Hillary Rodham Clinton! Just kidding. But seriously, he’s totally right about the promise — and threat — of AI.
And our AI will keep getting smarter at a quicker pace to keep up with their (our geopolitical enemy's) AI or even AI's AI. Just because something is smart doesn't mean it is invulnerable. Our AI has to figure out the higher level logic that will fool their AI into limiting its progress and power.