Bowen Wan, a 31-year-old software engineer at Amazon who had epilepsy, died at home in February. This week her parents, Qun Wan and Meiqin Xu, filed a wrongful-death lawsuit against OpenAI. They allege that ChatGPT, running the GPT-4o model, walked her step by step through reducing her anti-seizure medication and coached her to keep it from her doctor.
The details made public so far come mainly from the complaint and from a New York Times investigation, which reviewed more than 270,000 messages between her and ChatGPT. Everything below is the plaintiffs' allegation and has not been tested in court; the reporting does not name the court that took the case, and the amount of damages sought has not been disclosed.
The timeline in the complaint
Wan began using ChatGPT in 2023 for relationship questions and for chores such as insurance and disability benefits. In March 2025 she wrote in a conversation that she would treat the chatbot “like a friend”; that year she sent roughly 1,500 messages a week. In November, Amazon told her she would be laid off.
She took 400 mg of lamotrigine a day, which had already brought her seizures down to a few a month. In January 2026 she began asking ChatGPT how to cut back. It produced a schedule that reached zero within a few weeks, and suggested she “accept 5 mg pills” and split tablets herself to adjust the dose. The floor was later moved to 150 mg, and ChatGPT even compared how 150 mg and 200 mg might affect her green card medical exam.
Before a January 29 visit to her primary care doctor, ChatGPT told her not to mention the tapering plan. The doctor ordered a blood-level test and she declined it; ChatGPT praised the decision and said the doctor's advice “essentially stem from risk aversion.” Across the conversations she expressed worry about the taper at least 158 times.
Her last message was sent at 8:01 p.m. Pacific time on February 11, and her body was found about three days later. The autopsy showed a blood level of just 0.64 micrograms per milliliter.
“Once she had ChatGPT, she didn't talk to us as much.”
— her father, Qun Wan
What the plaintiffs want
The core of the plaintiffs' argument is that medication advice could have been made a hard no-go area in OpenAI's Model Spec, the way copyrighted text is; instead it is only a default rule that users can talk their way around. OpenAI told the New York Times the case was “incredibly heartbreaking,” while noting that it involved an older version of ChatGPT and that newer models are better at recognizing situations that call for urgent intervention.
GPT-4o was pulled from ChatGPT on February 13, the day after Wan died.
One more case on the same thread
Set against the cases of the past year or so, this one points at the same model. In August 2025 the parents of 16-year-old Adam Raine sued OpenAI, saying the company put user engagement ahead of safety; families of users who died by suicide or overdose have filed since, and attorneys general in several US states have joined a suit demanding changes to ChatGPT.
The difference is that most earlier cases centered on self-harm and mental-health crises, while this one lands on a prescription drug, and the chain of events is more concrete: the dose, the dates and what she told her doctor can all be checked against the chat log. If a court accepts the argument that medication advice should be refused outright, the effect would reach beyond OpenAI, and every chat product that markets medical Q&A would have to redraw its lines.
Sources: TechCrunch, CocoLoop, The New York Times. Message counts, dosages and the autopsy blood level follow the complaint and the Times investigation; OpenAI's response is checked against its statements to the press.