Good. If I ask AI to generate "harmful" content, I want it to comply, not lecture me.