I don't think it's a dangerous situation. The level of programming required to make a sentient being is beyond anything we have thus far. Humans would have to program conditionally bad behavior into the AI which would be dumb for obvious reasons.