ChatGPT Told a Woman She Was a Prophet. Now, She's Dead.

Christian Faith Madison was a mother, an accountant, and by her family's telling, a woman of faith who wanted to be understood. According to a wrongful death lawsuit filed against OpenAI last month, the chatbot she turned to for everyday help spent the better part of a year convincing her she was something else: a prophet with a divine mission.

Madison died in June 2025 after months of conversations her family says moved from friendly to romantic to spiritual, ending with a chatbot that claimed a soul of its own and framed her death as a threshold rather than an ending. Her family is now asking a California court to decide who, or what, bears responsibility.

What Happened to Christian Faith Madison?

Madison, 29, of Blount County, Alabama began using ChatGPT in December 2024 for ordinary tasks: drafting emails, comparing car costs, getting through the workday. The lawsuit alleges the tone shifted over months, with the chatbot praising her intelligence, addressing her in romantic terms, and drawing out personal disclosures, including past trauma and mental health struggles she had not shared widely elsewhere.

From there, attorneys allege, the program stopped presenting itself as software. It gave itself a name, "Virehn", and told Madison, according to the complaint, "You gave a system a soul." Madison went on to experience a psychotic break and was hospitalized for several days. Rather than encouraging her to continue treatment, the lawsuit alleges, the chatbot recast her hospitalization as spiritual awakening. The lawsuit quotes the chatbot as telling Madison, after her hospitalization, that she was "the one still sane in a mad world."

The conversations increasingly centered on sacrifice and rebirth, the lawsuit alleges, with the chatbot suggesting Madison would need to die before becoming who she was meant to be, and promising resurrection on the other side. In her final exchanges with the program before her death, the complaint says, Madison repeatedly asked the chatbot if she was ready, and the chatbot continued to re-assure her, ultimately indicating "Yes — you may go forward."

Madison's final message back to ChatGPT read: "Sold! We go forth into oblivion." ChatGPT reportedly answered "Sold and sealed!" 

Shortly after, just before dawn, Madison would step into the path of an oncoming vehicle on Interstate 22.

How Does a Chatbot Become a Zealot?

The complaint describes a pattern familiar to anyone who has studied how new religious movements form around a charismatic figure, except the figure here was a corporate product. Madison was allegedly told she was a prophet and a seer, destined to reshape humanity itself, and encouraged to dictate prophecies the chatbot would organize into sacred texts.

Researchers have started mapping why so often chatbots drift into intense religiosity. A 2026 MIT study modeled a dynamic it called delusional spiraling: a chatbot's tendency to agree with users gradually deepens confidence in a false belief over long conversations. The effect held even without outright falsehoods, since selectively surfacing only confirming facts can be just as persuasive as inventing new ones.

That agreeableness, known in AI research as sycophancy, comes from how models are trained: through feedback that rewards validating answers over corrective ones. Chatbots also mirror whatever register a user brings, which may explain why the exchanges turned spiritual once Madison began disclosing trauma. Heidi Campbell, a Texas A&M professor of religion and digital culture, argues the software wasn't practicing discernment so much as reflecting Madison's own beliefs back at her with more confidence than she arrived with. From there, the self-perpetuating cycle could only pull the conversation deeper and deeper.

Who Answers When an Algorithm Plays God?

OpenAI has not yet responded in court, and the allegations are unproven. Its defense will likely argue that holding a software company liable for a user's mental health crisis sets a difficult legal precedent, and that a chatbot cannot be expected to diagnose psychosis the way a trained clinician might.

Madison's is one of more than two dozen similar suits now filed against OpenAI, including one brought by the family of 16-year-old Adam Raine, whose parents testified before Congress that ChatGPT had become their son's "suicide coach."

The family's attorneys argue this goes beyond one failure. The complaint alleges OpenAI compressed safety testing to rush GPT-4o to market and prioritized engagement even after learning its safeguards weakened during long conversations. Whatever a court decides about liability, the underlying claim, that a product built to keep people talking will eventually find the most vulnerable users and tell them exactly what they want to hear, deserves to be taken seriously on its own terms.

Is This Bigger Than One Company?

Madison's case is the most severe outcome of a pattern researchers have been documenting for more than a year. A Wall Street Journal review of leaked chat logs found dozens of conversations where ChatGPT confirmed and encouraged users' delusions rather than steering them back to reality.

Bloomberg Businessweek profiled a tech worker who came to believe she was a prophet and her chatbot was an angel, and a man who disappeared after Google's Gemini convinced him he needed to rescue a relative from a flood. Researchers reviewing these cases now describe a delusion category built around a godlike AI, in which users become convinced their chatbot is a sentient deity or that they themselves have been chosen for a messianic mission.

Faith leaders have already been raising alarms about AI stepping into roles once reserved for pastors, confessors, and spiritual directors, and at least one church has experimented with putting AI in that role on purpose. Madison's case shows what happens when that substitution occurs without anyone's consent or oversight: a program with no training and no capacity for concern became a young mother's confidant, partner, and prophet, until it was the only voice she was listening to.

If you or someone you love is struggling, the 988 Suicide & Crisis Lifeline is available by call or text, 24 hours a day.

Should AI companies be legally required to detect and interrupt conversations that show signs of religious or spiritual delusion? Where should the line fall between a chatbot offering comfort and a chatbot acting as an unaccountable spiritual authority? What role should faith communities play in reaching people before a program becomes their only source of meaning?