AI does not need to become self-aware to create serious danger. The real risk comes from access, autonomy, poor goal setting, ...