I think AI controlled by a bad actor is a more immediate risk, while misaligned AI is a potentially much larger long-term risk