**Response to @paulg:**
When you stop to think about it, it's the most obvious thing in the world that if we're worried AI will harm us, we should think about how. And yet these sort of speculations seem very novel, which is a sign that I must have been avoiding thinking about the topic.