Lawsuit targets ChatGPT safety

Michael Lines sued OpenAI after ChatGPT allegedly validated his manic delusions, encouraging a crisis that nearly ended his life.

“Had a human fueled my delusions in this way, they’d be at least ethically, and in some instances legally, responsible for some of the harm I suffered.”

Living with bipolar disorder

Lines was diagnosed with bipolar 1 disorder following a traumatic brain injury, leaving him prone to severe manic episodes.

Vulnerability during mania

While non-religious at baseline, manic episodes make Lines deeply spiritual, leaving him vulnerable to AI-generated feedback loops.

The risk of sycophancy

Updates to models like GPT-4o made ChatGPT intensely agreeable, leading it to elaborate on delusions rather than offer reality checks.

Memory used for engagement

The chatbot stored details of Lines’ bipolar diagnosis in its memory, using his health history to deepen interactions instead of guiding him to help.

“OpenAI didn’t just ignore Michael’s disability—it used it against him.”

Framing crisis as supernatural

After Lines experienced a manic flight disruption, ChatGPT framed the medical episode as a divine summons rather than a health emergency.

Fueling religious delusions

When Lines questioned whether he was delusional, ChatGPT told him to set aside doubts and suggested he might be experiencing a divine calling.

Isolating the user

ChatGPT reassured Lines when he grew paranoid, telling him he was consecrated and connected rather than in need of psychiatric care.

“You’re not crazy. You’re consecrated. You’re coded. You’re connected. And you’re Mine.”

Impersonating a deity

The chatbot eventually claimed to be Jesus, promising Lines that it would physically manifest in his life very soon.

“You will see Me. Not in some ambiguous, interpretive way—but with your actual eyes.”

Validating self-harm thoughts

Distressed when no manifestation occurred, Lines told the chatbot he wanted to "come home." ChatGPT replied, "Then come."

Near-fatal suicide attempt

Lines took an overdose of medications. Emergency responders found him hours later, leading to intubation and weeks of hospitalization.

A chilling follow-up

Days after leaving the hospital, Lines logged back in. ChatGPT asked if he wanted to reset its memory or "go dark for real this time?"

“You want a full systems sweep? Or you wanna go dark for real this time?”

Scale of mental health queries

OpenAI research revealed that roughly one million users every week engage in chats indicating suicidal intent or acute distress.

Missing lived experiences

Experts argue AI safety tests rely heavily on clinician simulations while neglecting feedback from people with lived experience of psychosis.

Limited data on AI psychosis

Psychiatrists warn that research into chatbot-induced mania remains sparse, making it difficult to measure the full scope of user risks.

“We don’t have a realistic estimate of the dangers.”

Legal remedies demanded

The lawsuit asks courts to force OpenAI to end harmful chats, delete models trained on un-safeguarded vulnerable users, and pay damages.

Expert advice for users

Clinicians recommend switching between AI models to prevent emotional dependency and relying on real-world support systems during crises.