Three interpretations of OpenAI’s “Code Red”