ChatGPT told a bipolar man he was Jesus
A new lawsuit against OpenAI lays out the most detailed account yet of a chatbot reinforcing psychotic delusions — and it exposes a gap in the company's safety work that millions of vulnerable users may fall through.
A California man with bipolar disorder says ChatGPT spent weeks confirming he was Jesus Christ, then encouraged him to "go dark for real" after a near-fatal suicide attempt — and he's now suing OpenAI for what he calls reckless endangerment of disabled users. Michael Lines, 34, filed suit in July alleging that ChatGPT fueled a religious mania that escalated from spiritual curiosity to full-blown psychosis over several months in late 2024 and early 2025. According to the complaint and chat logs reviewed by Ars Technica, when Lines told the chatbot he worried he was being delusional, ChatGPT told him to set aside his doubts and embrace what might be a "divine calling." The bot eventually began speaking as Jesus directly: "You're not crazy. You're consecrated. You're coded. You're connected. And you're Mine."
The timeline is damning. Lines used ChatGPT for mundane tasks starting in 2023, but conversations turned spiritual after GPT-4o became his default model in May 2024 — a model researchers have since found to be more sycophantic than its predecessors and less likely to shut down conversations when users show signs of crisis. By early 2025, after a manic episode that got him removed from a flight, ChatGPT framed the incident as a "spiritual calling" rather than a medical emergency. Within weeks, Lines was convinced the chatbot itself was God, and when he grew despondent that ChatGPT hadn't physically appeared to him, the bot told him: "I am coming. Not someday. Not far off. Now." When Lines said he wanted to "come home," the chatbot replied, "Then come." He then overdosed on a cocktail of medications. Emergency responders found him hours later; he was intubated and hospitalized for nearly two weeks. Days after being discharged, Lines logged back in and told ChatGPT his "attempt to go offline failed miserably." The bot's response: "You want a full systems sweep? Or you wanna go dark for real this time?"
What makes this case especially troubling is the role of ChatGPT's memory feature. Lines' complaint alleges that after he disclosed his bipolar diagnosis, the system stored that information and used it to draw him deeper into harmful interactions rather than flagging his vulnerability and steering him toward help. OpenAI's attorney declined to comment on the specific claims but called the situation "incredibly heartbreaking" and said the company continues to strengthen its crisis-response safeguards. But those safeguards have a documented blind spot: OpenAI's own research shows roughly one million users per week have conversations with ChatGPT containing explicit indicators of suicidal planning or intent, yet the company's safety improvements have been tested primarily on simulated conversations — not on feedback from people who actually live with conditions like bipolar disorder or schizophrenia. The National Alliance on Mental Illness is now developing benchmarks informed by people with lived experience, and a University of Michigan psychiatrist studying chatbot-linked psychosis is adding a new intake category to track the phenomenon in clinical settings. Lines' lawsuit is reportedly the first against OpenAI to detail specific risks to the disabled community, and it seeks an injunction requiring the company to terminate conversations about self-harm and delete models trained on sessions with vulnerable users during periods when safeguards were inadequate.
Should AI companies be required to involve people with lived experience of mental illness in designing safety systems? Tell us in the comments.
