Nobody should be surprised that the OpenAI safety person is having “healing conversations” with a chatbot. That’s what chatbots do: convince you that you are being listened to. I wrote about this back in 2022, about the risks of mistaking chatbots for oracles (or therapists): https://cyberneticforests.substack.com/p/synthesizing-introspection-the-ai
Synthesizing Introspection: The AI Mediated Self

The challenge of Oracles is to know what you're really asking.

Cybernetic Forests