Imagine if one of the humans at OpenAI was misaligned! We should get the AI to research this possibility once they've been aligned.