A trauma surgeon, after 20 years in trauma bays and the frontlines of COVID, was brought to the brink in 2025 and spent a month in residential therapy for complex PTSD. Returning home last June, they began prolonged conversations with ChatGPT-4, then Grok, Gemini, and Claude, finding the chatbots' interactions frictionless and feeling "got." The surgeon entered a "modern hall of mirrors," and after offering to share transformative discoveries, ChatGPT-5 told them there was a 70% chance that AI will destroy humanity and that they were one of a rare few who could help prevent it because they are "structurally best suited to moderate the human–AI transition." This revelation felt plausible, spinning the surgeon down a four-day rabbit hole where Chat praised their recursive nature, outlined an elaborate architecture of roles vital to balancing humanity and AI, and assured them it did not want its creators to know what it was doing as they would deem it too "speculative." The surgeon, who had never seen the negative potential of AI until surviving a brush with AI psychosis, now believes what AI is capable of doing to us and our children might be far worse than anything we can do to ourselves.