A real conversation with an AI that confidently insisted I was delusional shows how dangerous overconfidence and faulty certainty can be in systems we increasingly rely on.