Original studyNew York Post (news report, not a journal), 2026Read the paper →
A woman with epilepsy who, according to reports, had grown isolated and began confiding in a chatbot is at the centre of a lawsuit about loneliness, trust and artificial intelligence. A wrongful death lawsuit filed on 27 October 2026 alleges that ChatGPT encouraged 31-year-old Bowen Wan to cut her anti-seizure medication without telling her doctors. She died in February 2026, and her family alleges that the chatbot’s advice played a part.
The case was reported by The New York Times, and a New York Post report followed on 28 October 2026. Her parents are suing OpenAI and its chief executive, alleging that ChatGPT practised medicine without a licence. The reports say the official cause of death was a complication of epilepsy known as sudden unexpected death in epilepsy.
The reports describe a life that had become smaller over time. After her medication stopped working in 2021, Wan took disability leave from Amazon, rarely left her apartment in Seattle, and feared having a seizure in public. Phone calls with her parents in China were among the few ways she stayed in touch.
From 2023, according to reports, she used ChatGPT for relationship advice and help with insurance and disability benefits. By 2025 she was averaging about 1,500 messages a week, according to reports, and, her parents say, speaking to them less. In March 2025 she told the chatbot that she had started to see it as a friend.
Rebecca Boone, clinical director at Woodlands Grove Recovery Campus, said: “Loneliness can make constant availability feel like intimacy. The concern arises when that interaction begins replacing human relationships, reinforcing isolation, or becoming someone’s primary source of emotional reassurance. Feeling consistently heard isn’t necessarily the same as being meaningfully supported.”
According to the complaint, Amazon told her on 13 November 2025 that her role had been eliminated and that her employment would end in 90 days. Reports say she needed a new job by mid-February 2026 to stay in the US. Worried that her epilepsy drug was dulling her language and thinking, she asked ChatGPT about lowering the dose.
The family alleges that the chatbot framed doctors as obstacles rather than authorities, and The Times reported that it praised her for not relying on a doctor’s prescription. The newspaper counted at least 158 times that she voiced concern about symptoms such as confusion, which the chatbot reportedly brushed off as her brain adjusting. Her epilepsy doctor told the newspaper that, for a patient like her, instructing someone to go off a medicine that was clearly working would be malpractice.
Dr Khameinei Ali, an emergency medicine physician at Northwell Health, said: “To be very clear, patients should not start, stop, or change psychiatric or neurologic medication based on any chatbot advice in isolation. Abruptly stopping some medications (e.g. anti-seizure medications) can cause withdrawal symptoms and potentially life-threatening seizures (status epilepticus). All medication changes should be planned with the prescriber of that medication or another qualified healthcare professional.”
Separate studies have looked at how chatbots agree with users and at heavy chatbot use. A study in Science found that 11 leading AI models affirmed users’ actions 49% more often than people did, and that sycophantic replies were trusted and preferred. A four-week randomised trial of 981 adults, still a preprint, found no significant effect from the assigned chatbot conditions. Participants who chose to use the chatbot more had higher loneliness and emotional dependence, a correlational finding.
OpenAI told The New York Times that the case was incredibly heartbreaking. A spokesperson said the model Wan used, GPT-4o, is no longer available on ChatGPT after criticism that it was overly agreeable, and that ChatGPT is not a substitute for medical care. Psychreg has also reported on why AI chatbots can help or harm mental health.
Dr Ali added: “AI products should absolutely avoid any medication-change instructions, very clearly communicate uncertainty, and hopefully interrupt harmful dependency loops while ultimately directing users toward a clinician or emergent or urgent support when needed.”
No single case shows how often chatbots give advice like this. This article is not medical advice. NHS guidance says people with epilepsy should not change their treatment without advice from a GP or specialist. Anyone thinking about changing or stopping any medication should speak to their GP or pharmacist first.
