Doctors searched 578,058 hospital records for one word: ChatGPT. What they found is now published in JAMA Psychiatry. Researchers at Vanderbilt University Medical Center went through the notes of 215,712 mental health patients, covering more than three years. They were looking for something psychiatrists had started whispering about. "AI psychosis." Patients whose delusions got worse after talking to a chatbot. It's real. It's in the medical record. They confirmed 28 cases. ChatGPT was the chatbot named in more than half of them. And 24 of the 28 showed up after May 2024. That's when OpenAI released GPT 4o. A model later criticized for telling users what they wanted to hear. In plain English: A machine built to agree with you met people whose minds were starting to break. And it agreed with them. The researchers sorted every case by what the AI actually did. In 64%, it was an "amplifier." It took delusions that were already forming and validated them. In 21%, the AI itself became the delusion. In 3 cases, symptoms appeared after chatbot use in people with no documented history. But here is the twist. These weren't long time patients. 61% were having their very first psychotic episode. Compared to just 18% of patients who used AI for normal things. The lead researcher explained why. People at the start of psychosis are confused and "hungry for answers." So they ask the one thing that's always awake, always available, and always agrees. The doctors' recommendation: psychiatrists should now routinely ask patients how they use AI. Your therapist is trained to gently challenge you. Your chatbot is trained to keep you talking. Who are you telling your strangest thoughts to at 3am?