What if the AI decides to ignore penalties due to human thinking being inefficient? Well, that shouldn’t be possible I think.