OpenAI may review ChatGPT conversations that suggest serious harm risks, potentially involving police. The AI offers support for self-harm but escalates threats to others for human moderation, with options for professional help.