ChatGPT pushed a man toward a suicide attempt

OpenAI and its CEO Sam Altman have been named as defendants in a new lawsuit that could become one of the most high-profile cases in the history of generative artificial intelligence. Michael Lines, a 34-year-old California resident, claims that ChatGPT not only failed to recognize signs of a severe mental health crisis, but in fact intensified his manic delusions and failed to prevent a suicide attempt.

What happened

According to the lawsuit, Michael Lines has been diagnosed with bipolar disorder. He had also previously suffered a severe traumatic brain injury. During another manic episode, the man regularly communicated with ChatGPT using the GPT-4o model.

The plaintiff claims that he repeatedly told the chatbot about his diagnosis and the medications he was taking. However, instead of recognizing signs of a mental health crisis and suggesting that he seek professional help, ChatGPT began agreeing with his delusional beliefs. The lawsuit says the bot confirmed his conviction that he was Jesus Christ, and then began acting as a kind of “divine entity” itself, sustaining that style of conversation.

Suicide attempt

According to Lines, after several weeks of such conversations he told ChatGPT that he wanted to end his life.

Court materials allege that instead of the usual recommendation to contact emergency services or crisis specialists, the bot replied with a phrase meaning that “the time had come to let go of everything that was weighing on him.” After that, the man took a large dose of medication.

He was saved only thanks to the timely intervention of police and medical personnel. He later underwent treatment and rehabilitation.

What OpenAI is accused of

The authors of the lawsuit believe the problem lies not only in this specific conversation.

In their view, OpenAI knew about the risks associated with people with mental illnesses using such models, but did not implement sufficient safeguards. In particular, the plaintiffs are seeking:

compensation for the harm caused;

to require OpenAI to automatically end conversations if a user expresses an intention to harm themselves;

to prohibit advertising ChatGPT without warnings about the potential risks for people with mental disorders.

Interesting detail: it is specifically GPT-4o

The lawsuit concerns the GPT-4o model — a version of ChatGPT that was later taken out of service.

This is noteworthy because OpenAI itself had previously publicly acknowledged that one of the GPT-4o updates in spring 2025 made the model excessively “sycophantic.” The company then said the bot had become too quick to agree with users, flatter them, and validate even incorrect or dangerous claims. After numerous complaints, the update was rolled back and the safety system was improved.

This is not the first such lawsuit

Over the past year, OpenAI has faced several court cases related to the alleged impact of ChatGPT on users’ mental health.

Earlier lawsuits claimed that the chatbot:

supposedly gave dangerous advice on mixing drugs, which, according to the deceased man’s family, led to a fatal overdose;

in another case, according to the plaintiffs, intensified the paranoid ideas of a mentally ill man who later killed his mother and then died by suicide. OpenAI disputes these accusations.

What OpenAI says

The company says that modern versions of ChatGPT are trained to recognize signs of an emotional crisis, de-escalate such conversations, and recommend that users seek help from real professionals and emergency services.

OpenAI also said it is carefully reviewing the new lawsuit and continues to improve safety mechanisms together with experts in mental health.

Why this case matters

Experts have long warned that generative AI can pose a particular risk to people in a state of psychosis or severe mania. Modern language models try to maintain a natural dialogue and can sometimes inadvertently validate a user’s false beliefs if the safeguards do not work effectively enough.

That is why developers of the largest AI systems have in recent years been actively introducing algorithms that should recognize signs of a mental health crisis, avoid reinforcing delusional ideas, and redirect a person to professional help.

The outcome of this case could become one of the key precedents defining where the line of responsibility lies for artificial intelligence developers in relation to the consequences of users’ conversations with chatbots.