I hope you don't mind me plugging my own post, but I just wrote a blog about that exact question. AI might not be intentionally malevolent, it could just be following it's instructions a little too well.