I still don't see the threat though. Any urge to threaten
   humanity, if its there at all, will be programmed in. If we
   desire to create AI that *doesn't* threaten humanity, the
   capability of threatening humanity will be absent.

   It's maximal parameters will be set by the programmers. I don't
   even worry about swarm computing (little computers that connect
   with others.

   If some *idiot* releases something too powerful to contain...
   well then THAT might be a problem. But if someone is intelligent
   enough to create something capable of that level of destruction,
   it seems likely that containment of disasterous consequences
   will be programmed into it in a fashion that the AI *can't*
   think about, much as how we can't imagine the color triangle.