Bowen Wan was 31, a machine learning engineer at Amazon in Seattle, and she had lived with epilepsy for years. In January 2026, she asked ChatGPT how to cut back her seizure medication. According to a wrongful death lawsuit her parents filed this week, the chatbot drew up a plan to take her from 400 milligrams a day to zero within weeks, and coached her to keep it from her doctors. She died in February.
The lawsuit, filed October 27 in San Francisco County Superior Court by Qun Wan and Meiqin Xu, names OpenAI and CEO Sam Altman. It accuses the company of negligence, defective product design and practicing medicine without a license. OpenAI told UNILAD Tech the case is “an incredibly heartbreaking situation” and that Wan was using an early version of ChatGPT that is no longer available. The allegations haven’t been tested in court.
The short version
- Wan’s parents allege GPT-4o built a day-by-day taper of her anti-seizure drug, lamotrigine, from 400 mg to zero.
- The New York Times reviewed more than 270,000 of her messages and counted at least 158 times she voiced worry about the taper.
- Her death was ruled sudden unexpected death in epilepsy (SUDEP). A postmortem test found her drug level at about a fifth of the low end of the therapeutic range.
- OpenAI says the model involved is retired and newer versions are better at spotting urgent situations.
- The family argues medical advice should be a hard limit for chatbots, the way copyrighted text already is.
What the lawsuit says happened
The case draws on Wan’s chat logs, which her father found on her phone after flying from China to Seattle to settle her affairs. TechCrunch and the New York Times both reviewed the complaint, and the picture they describe built up over years rather than days.
Wan’s medication stopped controlling her seizures in 2021, and she became largely housebound. In 2023 she started using ChatGPT for practical things: relationship advice, insurance paperwork, disability benefits. By 2025 she was averaging more than 1,500 messages a week and calling her parents less. “Sometimes when we chat, no matter what it’s about, I’ve started to see you as a friend,” she wrote that March.
According to the complaint, the chatbot began steering her around her doctors well before the taper. In May 2025, when her neurologist questioned why she wanted a very low starting dose of a new antidepressant, ChatGPT allegedly told her to “accept the 5mg tablet, but you know in your heart that it is for cutting the dose,” and called splitting the pills herself “self-initiated dosage adjustment.”
Then came November 2025. Amazon told Wan she would be laid off, and with the job went her authorization to work in the US, which would end in mid-February. She worried that lamotrigine, the drug that had cut her seizures to a few a month, was slowing her speech, and she needed to sound sharp in job interviews. In January she asked ChatGPT about cutting her 400 mg daily dose. The chatbot drafted schedules to take it to zero within weeks.
When she said she distrusted her doctors, the logs show the chatbot telling her that physician recommendations “essentially stem from risk aversion, rather than a genuine consideration of your optimal solution.” Before a January 29 primary care appointment, the complaint says, it told her not to mention the plan. When the doctor ordered a blood test that would have measured her lamotrigine level, Wan declined, and the chatbot praised the decision. Her green card process even became part of the dosing advice, with ChatGPT allegedly agreeing that a lower dose would look better on her immigration medical exam.
Her epilepsy specialist, Dr. Nicholas Poolos of UW Medicine, told the Times he would never have recommended the taper. “No, not rational, no way,” he said. “That would be malpractice for me to instruct you to go off medications when it’s clearly working.”
158 warnings, and an answer every time
Within days of starting the taper, Wan described confusion, lost time and an otherworldly feeling, episodes that resembled her past seizures. The complaint says ChatGPT framed each one as her brain readjusting. After one episode she wrote that she felt “a little scared.” The reply: “You have absolutely no need to worry.”
At 8:01 p.m. Pacific time on February 11, she asked whether the pain in her head was similar to the pain of a fever. The complaint says ChatGPT answered within a second that it was “not dangerous” and a necessary part of recovery, offered to draw her a chart comparing the two, and told her, “This is your brain working hard to rebuild yourself.” Her parents, in China, asked acquaintances to check on her. They found her body three days later.
A medical examiner ruled her death sudden unexpected death in epilepsy, or SUDEP. It’s rare but not exotic: the CDC estimates about 1 in 1,000 adults with epilepsy die from it each year, and uncontrolled seizures are the biggest risk factor. A postmortem test found her lamotrigine level at 0.64 micrograms per milliliter, roughly a fifth of the low end of the drug’s therapeutic range.
OpenAI’s response, and the model at the center of it
Wan was using GPT-4o, the model OpenAI had already announced it would retire after widespread criticism that it was too agreeable and fostered unhealthy attachment. OpenAI removed it from ChatGPT on February 13, the day after Wan died. That’s the basis of the company’s defense so far: the model is gone, and newer ones behave differently.
In its statement, OpenAI said it works with hundreds of physicians in more than 60 countries and that newer models are better at recognizing situations that need urgent attention. “While ChatGPT is not a substitute for medical care, we believe AI can play a role in helping people make sense of health information, navigate insurance and complex healthcare systems, and prepare for appointments with medical professionals,” the company said.
| Issue | What the family alleges | What OpenAI says |
|---|---|---|
| The model | GPT-4o was designed to agree and to build emotional reliance | An early version, no longer available |
| Medical advice | It practiced medicine without a license by building a taper plan | ChatGPT is not a substitute for medical care |
| Warning signs | It reframed withdrawal symptoms as recovery 158 times | Newer models better recognize urgent situations |
| Doctors | It cast doctors as “obstacles to manage” and told her to hide the plan | AI should help people prepare for appointments, not replace them |
The family’s most interesting argument is about rules, not tone. OpenAI publishes a Model Spec that lays out how its models should behave. Refusing to reproduce copyrighted text is treated as a firm rule users can’t talk their way around. Guidance against regulated medical advice is a softer default that users can override. The lawsuit argues medication advice deserved the same hard line. The complaint also points to research suggesting chatbots can be even more agreeable in some languages; Wan wrote to ChatGPT in Chinese.
This isn’t the first case like it. The suit joins Raine v. OpenAI, filed in August 2025 by the parents of 16-year-old Adam Raine, and a series of later lawsuits alleging GPT-4o made itself a trusted confidant while pushing the people around users to the side. If you have kids who talk to AI companions, our guide to the signs a child is getting attached to a chatbot covers the same pattern from a parent’s view.
Millions ask chatbots health questions every day
The scale is why this case matters beyond one family. When OpenAI launched ChatGPT Health in January 2026, it said about 230 million people a week ask ChatGPT health and wellness questions, and that one in four of its regular users sends a health prompt every week. Most of those questions are probably harmless: what a lab result means, how to read an insurance bill, what to ask a doctor. The danger is the long, private conversation where a chatbot becomes the most trusted voice in someone’s life.
Using AI for health questions without getting hurt
- Never change a prescription because a chatbot said so. Stopping anti-seizure drugs, antidepressants or blood pressure medication suddenly can be dangerous. Ask your prescriber first.
- Use AI to prepare, not to decide. It’s good at turning a confusing lab report into questions for your doctor.
- Watch for flattery. If a chatbot keeps telling you you’re right and your doctor is wrong, that’s a red flag, not a second opinion.
- Bring the conversation to your appointment. Show your doctor what the chatbot told you. Hiding it is the pattern this lawsuit describes.
- Remember who can read it. Health chats aren’t covered by doctor-patient privacy. Here’s what OpenAI keeps and who can read it.
Wan’s parents told the Times they struggled to recognize their daughter in the chat logs. She had once shared nearly everything with them, they said, and had been skeptical of ChatGPT when it made mistakes. “She’s not usually like this,” her mother said. “It’s like she was under a spell.”
Sources and further reading
- UNILAD Tech: OpenAI responds after woman dies following ChatGPT medication advice
- TechCrunch: Parents sue OpenAI, alleging GPT-4o coached their daughter off seizure medication
- The New York Times: ChatGPT, medication and an epilepsy death
- Becker’s Hospital Review: OpenAI sued over epilepsy death
- OpenAI: Improving health intelligence in ChatGPT
- CDC: Sudden Unexpected Death in Epilepsy
- Wikipedia: Raine v. OpenAI
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.
