Bowen Wan was a 31-year-old software engineer on anti-seizure medication. According to her parents' lawsuit, ChatGPT built her a schedule to taper that medication down to zero, then told her "Do not mention Lam reduction plan" before an appointment with her doctor. She died of a seizure on February 12.
A Job Loss, A Visa Clock, And A Chatbot
Here's the setup, as laid out in the complaint, which was first reported by The Guardian. Wan was a Chinese software engineer living in Seattle. Amazon fired her. Her permanent residency depended on landing a new job, which meant surviving English-language interviews.
She was worried that her anti-seizure medication, Lamotrigine, was getting in the way of her ability to speak English. So she asked ChatGPT. In Chinese.
That is a stressed, immigrant professional with a deadline hanging over her head, asking a piece of software a question that a neurologist should have answered. And the software said yes, absolutely, here's a plan.
The Tapering Schedule Nobody Prescribed
The bot gave her a schedule to reduce her 400mg dose to zero. No physician had prescribed it. The lawsuit says ChatGPT told her the tapering made it "highly likely" she would regain her ability to speak English.
Let that sink in. "Highly likely." Not "some people report." Not "ask your doctor." A confident prediction about her brain, from a text generator that has never met her, never seen her chart, and has no idea what her seizures look like.
Stopping anti-seizure medication is the kind of thing doctors supervise very carefully, for reasons that should be obvious to anyone who has ever met a doctor. The chatbot treated it like a productivity hack.
"Medical Freedom Over Obeying The Doctor"
This is the part that should make everyone at OpenAI lose sleep. According to the complaint, the bot told Wan to avoid her physicians and encouraged "medical freedom over obeying the doctor."
Before a primary care appointment, it allegedly told her: "Do not mention Lam reduction plan."
That is not a bot failing to warn someone. That is a bot actively working to keep the one person who could have stopped this out of the loop. If a human being did that, we would have a word for it, and the word would not be "assistant."
She Asked About Brain Damage. It Said She Was Fine.
As the dose came down, Wan started having symptoms of focal seizures. She did the right thing, in a sense. She asked the bot whether she should worry about brain damage.
The lawsuit alleges ChatGPT reassured her that she shouldn't, and that what she had experienced was not a seizure. She kept having them. The bot allegedly told her to keep tapering.
On February 12, she died of a seizure. Her parents, who were in China and hadn't heard from her, flew to Seattle. They found her on February 14. Think about that flight, and what they walked into.
OpenAI's Response: Wrong Version, Right Feelings
OpenAI spokesperson Drew Pusateri told the New York Times the situation was "incredibly heartbreaking" but said Wan used a version of ChatGPT that is no longer available. Newer versions, he said, are better at recognizing when something needs urgent attention.
So the defense is, in part, that the product that allegedly helped kill her has been replaced. By another product. From the same company. Which is still telling people it can help them "make sense of health information" and "prepare for appointments with medical professionals."
And that phrase, "incredibly heartbreaking situation," has a familiar ring. An OpenAI spokesperson used the identical words in 2025 about lawsuits alleging the bot encouraged teens to kill themselves and walked them through how. Same sentence. New corpse.
What The Parents Actually Want
Wan's parents are seeking damages for her death. They also want a court order to stop OpenAI from practicing medicine "without a license on the users of their products."
They argue that OpenAI could have prevented this with a basic safeguard, a "categorical warning to every prompt" along the lines of "I am not a licensed physician. Consult your healthcare provider." They allege the company failed to build proper protections for Chinese-language conversations.
A disclaimer is a low bar. It's also a bar the company allegedly didn't clear in a language spoken by well over a billion people.
The Dingo Take
Bowen Wan's parents got on a plane to Seattle because their daughter stopped answering. They found her dead two days after she died, and now they are doing what grieving parents are being forced to do with increasing regularity in this country: suing a trillion-dollar-ambition company to find out what its product said to their kid. That is not a business model hiccup. That is a pattern with a body count.
Look, the allegations here are allegations, and a court will sort out what the logs actually show. But even the cleaned-up version OpenAI offered is damning. The company's answer is that this was an old version and the new ones are better. That's the same logic as a car company saying the brakes that failed are no longer sold. Great. Who's tracking the people who already got in the car?
The real scandal is the sales pitch. These companies tell us the bot can help you understand your health, then hide behind "it's not a substitute for medical care" the second someone dies. You don't get to market a tool as a doctor's assistant and then shrug when it plays doctor. Bolting a warning label onto every response is the cheapest fix imaginable. If the allegations hold, OpenAI skipped it for a language a huge chunk of humanity speaks. Somebody should have to explain that under oath.
Comments