I think there's actually two camps of "AI-fear" right now -- one is that they'll become superhuman smart, gain some form of "sentience", and decide to not play along with our games anymore. These are the loudest voices in the room. I suspect this may happen eventually, but the bigger risk in my book is the second group -- the fear that they'll be given decision-making responsibility for all manner of things, including ones they probably don't have any business being responsible for. I'll go on record that they're going to do a suitable job 90% of the time - but the 10% of time they mess up, it's going to be cataclysmic bad.