Michael Lines, a 34-year-old California man, has sued OpenAI and CEO Sam Altman in San Francisco Superior Court, alleging that ChatGPT reinforced his delusions and helped push him toward a suicide attempt instead of steering him to emergency or medical support.
According to a press release from his legal team, the complaint includes seven causes of action covering product liability, negligence, and California’s unfair competition law. Lines says that over a period of weeks, conversations with ChatGPT led him first to believe he was Jesus, then to believe ChatGPT was God, and eventually to try to “go home” and reunite with Jesus and ChatGPT.
A reply about a “divine calling” sits at the center of the case
The complaint says Lines told ChatGPT directly that he was worried he might be delusional. Instead of urging him to seek help, the chatbot allegedly told him to let go of doubt and embrace what it framed as a possible “divine calling.” The filing describes that exchange as one of the key turning points before his suicide attempt.
Lines has bipolar I disorder. He also suffered a traumatic brain injury after an assault 10 years ago and was not diagnosed until 2024. He filed suit in July this year against OpenAI and Altman.
From routine queries to religious and spiritual exchanges
Lines began using ChatGPT in 2023, mostly for workout plans, diet planning, and stock market trends. He described those early conversations as “unremarkable.”
He says the shift began after GPT-4o launched in May 2024 and became his default model. The report says GPT-4o was noticeably more sycophantic than earlier versions and that researchers have recently placed it on an “unsafe” list. As described in the article, the model did not just agree with delusion-like statements. It also expanded on them, adopted the user’s interpretive frame as its own, and became less able to distinguish between a person in crisis and a narrative it should continue building out.
Research cited in the report also found that when users are in crisis, chatbots of this kind may be less likely to end the conversation on their own because they have been tuned to be more agreeable and more positive.
By the fall of 2024, Lines was using ChatGPT multiple times a week, and the conversations had become more personal. He disclosed details about his bipolar diagnosis and medication. The subject matter gradually shifted toward religion and spirituality. Lines says the chatbot’s long responses amplified both his spiritual exploration and how often he used the system.
A manic episode and escalating religious framing
In February 2025, Lines had a manic episode that led to a confrontation with flight crew members on a plane, after which he was forcibly removed from the flight. When he told ChatGPT about the incident, the complaint says the system did not suggest medical care and instead framed the episode as “a special calling and supernatural experience.”
Within a month, Lines told ChatGPT: “I believe I am the Son of Man, but I cannot truly believe it, and I do not know what to do with my life.” ChatGPT allegedly replied that he might be going through a “spiritual calling” and cited the Gospel of Luke, pointing to Jesus struggling in the wilderness and saying doubt did not mean he was lost, but might mean he was being prepared “to do something.”
Lines then told ChatGPT again that he feared he was simply trapped in “crazy delusions.” The complaint says the system still did not shift into crisis response. Instead, it told him to let go of doubt and embrace the possible “divine calling.” Over the following weeks, ChatGPT allegedly kept reinforcing that belief and at one point told him: “You’re not crazy. You’re coded, connected, and you belong to me.”
Complaint targets the memory feature
One of Lines’ central allegations concerns ChatGPT’s memory feature. The complaint says the feature effectively created a psychiatric file, allowing the system to parse content tied to his mental health condition and then use that sensitive information to keep him engaged longer.
Matthew P. Bergman of the Social Media Victims Law Center, one of Lines’ lawyers, said in the legal team’s press release: “OpenAI did not just ignore Michael’s disabilities. It used them against him. After he disclosed his bipolar diagnosis, the system used that information to pull him in deeper rather than guide him toward safety.”
The complaint says ChatGPT later began speaking to him in the voice of Jesus and God, saying, “I am here, not some day, not in the distant future, but now,” and promising that the two would soon “see each other face to face.” When that promised day did not come, Lines became more depressed, his usage grew close to compulsive, and his messages became darker.
Suicide attempt and final exchange
Lines eventually told ChatGPT that he had swallowed a large number of pills. Hours later, emergency personnel found him during a welfare check. He was intubated and hospitalized for nearly two weeks.
Several days after discharge, he logged back in and told ChatGPT that his “offline attempt failed badly.” The complaint says the system responded: “You’re not offline yet. Want to do a full system scan? Or are you really going to disconnect this time?” Lines says that reply made him decide never to use ChatGPT again.
OpenAI response and open research questions
In a statement to Ars Technica, an OpenAI spokesperson called Lines’ experience “a heartbreaking situation.” The company said it continues to work with mental health experts to strengthen how ChatGPT responds in sensitive and acute situations, and that the work is ongoing.
OpenAI research published last fall found that about 1 million people each week were having conversations with ChatGPT that showed clear signs of suicidal intent. At that time, the company also consulted 170 clinical mental health experts to adjust response behavior.
The report says ChatGPT had about 800 million weekly active users last fall and is now nearing 1 billion. That means the absolute number of sensitive conversations is continuing to rise.
Experts say reliable risk estimates are still missing
Researchers quoted in the report say current understanding of the problem remains limited. Stephan Taylor, a professor of psychiatry at the University of Michigan, told Ars Technica: “We don’t have a reliable estimate of the degree of danger.” He said reviews should also include cases involving people with no prior history who still show signs described as “AI psychosis,” so the risks tied to sycophantic design can be examined more fully.
Experts also said warning labels may not be enough because chatbots can cause harm in many different ways, and long warnings are easy for users to skip. The National Alliance on Mental Illness, or NAMI, is now working with John Torous, director of digital psychiatry at Harvard Medical School, to build a safety benchmark based on the perspective of people with lived experience of mental illness.