In any case even AI completely indistinguishable from ourselves is likely only to be as much of a threat as our children are. Everything depends on how it is managed. Apocalptic scenarios generally can only happen in a tightly connected, online society with no offline fallback mechanisms for control. If we are that careless/stupid, then perhaps AI taking over our systems (and hence us) is not so bad after all...?
AI needs creative intelligence, ability to learn and adapt the very process by which it learns, as well as links to the physical world enabling it to control and perpetuate its existence. The entire process is in our (human) control. Until the very end.
The other direction is frankly more scary. Augmented human beings and other existing living beings. That is much more prone to go haywire and pose a threat.