Yes indeed. But an AI isn’t needed to exploit one. I’d be more worried about a malicious individual utilizing social engineering exploit exploit than I would an AI system magically going rogue and doing so. And it would require some human involvement to train such a system to begin with. And then give it access to an intended victim.
Being “smart” isn’t the same thing as being “intelligent” or “sentient.” A so-called smart bomb can wreak all sorts of havoc. But that doesn’t mean it’s sentient. It still needs someone to make the decision for it to be deployed. And that decision itself is a type of on/off switch as well. And in the case of potentially dangerous that can be executed by highly automated systems there are multiple failsafes and levels of authorization required to enable these systems for certain actions.
I worry less about AI technology and more about increasingly powerful computing platforms that could render our current encryption technology useless. A disturbing thought when you consider everything from e-commerce to the systems that manage our infrastructure and defense systems all rely on encryption to make them secure. We’re developing systems that have the horsepower to steamroll over encryption which has remained secure mostly because the complexity of cracking it was beyond the capabilities the Von-Neumann architecture most computers are currently based on. However, new computer architectures that go beyond our current supercomputers capabilities are on the near horizon. And experimental systems that utilize them already exist.
