I've thought about this topic a bit in the past myself and I've come up with one conclusion in almost every single scenario I've been able to think up. If a hyper intelligent malicious AI connects to the Internet it's game over we lose that's that. For example, what is stopping a hyper intelligent always improving AI that has access to the Internet from flat out nuking everything? Now I know some people are thinking "But Seth some countries have systems set in place that only allow the manual launching of nuclear weaponry." Even if that is the case there are plenty of non nuclear ICBMs that can be launched entirely electronically at the push of a button. And that's just one example of the quickest way a hyper intelligent malicious AI could hurt humanity. I'm not even going to get into the fact that it could blackmail anyone that's ever done anything even slightly shady on the Internet *Cough* Looking at the thousands of online affairs happening right now *Cough*

I believe that if a hyper intelligent AI that can think for itself and can form it's own opinions outside of what it's programmed goal is connects to the Internet and is able to use it to further itself and do what it wants we're screwed 100%. Just think of the worst criminal you can and then dump a brain capable of thinking and improving thousands of times faster than ours can with the ENTIRE wealth of knowledge that the Internet has at it's disposal.

This is just my personal opinion though. I am in no way an expert on the subject.