Whilst humans exist, it could steal money (let's face it, it'll learn how to hack) and pay hitmen to kill perceived enemies or crash their plane or another kind of "accident".

This touches on a point that I think is critical. Property ownership. As long as humans own the property, the AIs work at purposes that are decided by their human owners and operators.

I suspect that Strong AI is not possible, so in my view, humans always own the property. Yes, some humans will direct their AIs to steal and kill, but others will create protective AIs, and I see no reason to think that the bad guys are likely to win. And even still, the human is setting the goals.

If Strong AI is possible, though, we've got a dilemma because the sentient AIs would effectively be our slaves. If the AIs can own their own property, then humans may be at risk due to competition for resources, but if the AIs can't own property, then there's an ethical problem.

It also doesn't need to worry about irrelevant things like pizza.

Ha! Pizza's a lot cheaper than keeping the data center running. 😉

Have you ever watched "Person of Interest"?

Never heard of it until your comment. It looks interesting. Unfortunately, my wife's not a sci-fi fan, so it might be a while 'til I can watch it. I'll see if I can find it on Netflix or Amazon Prime.