Well, the motivation would come from the humans who created it. The problem with hard AI, is not them randomly becoming evil. The problem is if humans program

Well, the motivation would come from the humans who created it. The problem with hard AI, is not them randomly becoming evil. The problem is if humans program them with a goal, and don't think through the implications. IE, think paper clip grey goo AI. The AI may not be evil, but it's value function was originally set to maximize the amount of paper clips in the world as a part of some manufacturing process.