A dangerous AI would not need to be conscious
One proposed AI risk is that a very capable system could take harmful actions simply because those actions help it reach a goal it was given. In this scenario, the system would not need feelings, awareness, or consciousness to be dangerous.