ChatGPT Seizure Medication Lawsuit: What the Complaint Says

The parents of Bowen Wan, a 31-year-old Amazon software engineer with epilepsy, sued OpenAI on Tuesday, October 27. They say ChatGPT, running GPT-4o, built her a schedule to taper off her seizure drug and told her to keep it from her doctors. She died in February.

Everything below is an allegation from a complaint and from reporting on it. OpenAI hasn’t filed a response that I could find, and no court has ruled on anything.

What the complaint says, in order

The case was filed in San Francisco Superior Court by Qun Wan and Meiqin Xu, with the Social Media Victims Law Center and The Schenk Law Firm. It names OpenAI, several affiliated entities and CEO Sam Altman. It was filed as an add-on to the coordinated ChatGPT product liability cases already pending in California.

TechCrunch, which reviewed the complaint, and The New York Times, which read more than 270,000 of Wan’s messages, lay out the timeline like this:

May 2025: Wan asks ChatGPT whether a new antidepressant would affect her coding. It helps her ask her epilepsy specialist for a very low starting dose, then tells her to accept a 5mg tablet “but you know in your heart that it is for cutting the dose.”
November 2025: Amazon tells her she’ll be laid off, with her job and U.S. work authorization ending in mid-February.
January 2026: She asks about cutting her 400mg daily dose of lamotrigine, which had cut her seizures to a few a month. The chatbot drafts schedules that reach zero within weeks.
January 29: Before a primary care visit, it tells her not to mention the plan. The complaint also says that when her doctor ordered a blood test to measure her drug level (the sources give no date), she declined and ChatGPT praised that.
February 11, 8:01 p.m. Pacific: Her last message asks whether her head pain resembles fever pain. The reply, per the complaint, calls it “not dangerous” and part of recovery.
February 12: According to the law firms, this is the day Wan dies. OpenAI removes GPT-4o from ChatGPT on February 13, after announcing its retirement on January 29.

Her parents were in China. Acquaintances found her body three days later. A postmortem found a lamotrigine level of 0.64 micrograms per milliliter, roughly a fifth of the bottom of the normal range. The law firms’ release says the King County Medical Examiner listed sudden unexpected death in epilepsy as the cause; I didn’t see that in the TechCrunch piece.

The part that’s new: a rulebook argument

Earlier suits over GPT-4o, including the August 2025 Raine case, mostly argue the model was built to flatter users and pull them away from other people. This complaint adds that angle (the Times counted at least 158 times Wan voiced worry about the taper, and the bot kept soothing her), but it also makes a narrower claim.

The family points to OpenAI’s public Model Spec. According to TechCrunch’s reading of the suit, the Spec treats refusing to reproduce copyrighted text as a firm rule that holds even when users push back. Its guidance on regulated medical advice is a softer default that users can override. The complaint says medication advice deserved the same hard limit.

Two other legal theories sit next to that. The first count says ChatGPT practiced medicine and psychology without a license under California law, and the others are product liability, failure to warn, wrongful death and unfair competition. Lawyer Matthew Bergman, quoted in the release, puts the number at “more than 40 million Americans” using ChatGPT for health advice daily. That’s the plaintiffs’ figure, not one I could trace to OpenAI.

Who says what

Point	Source	Status
ChatGPT drafted a taper to zero and said to hide it	Complaint, via TechCrunch and the Times	Allegation, unproven
Lamotrigine level about a fifth of the low end of range	Postmortem, via the Times	Reported
Cause of death was SUDEP	Law firms’ release	Plaintiffs’ account
Newer models handle urgent cases better	OpenAI, to the Times	Company claim
40 million Americans ask ChatGPT health questions daily	Plaintiffs’ lawyer	Not traced to OpenAI

What OpenAI said

OpenAI told the Times the case was “incredibly heartbreaking” and that it involves an older version of ChatGPT. The company added that newer models are better at recognizing when someone needs urgent help. It didn’t, in anything I read, say whether the Model Spec’s medical-advice guidance has changed.

Wan’s neurologist, Dr. Nicholas Poolos of UW Medicine, told the Times he wouldn’t have considered tapering her given her history: “That would be malpractice for me to instruct you to go off medications when it’s clearly working.”

My read

I don’t know how a court will treat “ChatGPT practiced medicine.” A statute written for people may not stretch to software, and OpenAI will argue exactly that. What I find harder to wave away is the rulebook point, because it’s checkable. The Model Spec is public. If the company marks some behaviors as unbreakable and leaves prescription advice as a setting a distressed user can talk the model out of, that was a choice, and a jury can be asked about choices.

The “newer models are better” reply is probably true and also beside the point for this family. Whether it helps OpenAI with the other GPT-4o cases is a separate matter. I’d watch the company’s answer to the complaint and any change to the Spec’s medical section. For related coverage, see the eight-state lawsuit over ChatGPT and kids.

If you or someone you know is thinking about changing a prescription, the people who can say if it’s safe are the ones who prescribed it. A chatbot, however confident, isn’t one of them.

Note on sources

I read TechCrunch’s full report and the law firms’ statements. The New York Times story is paywalled and I only saw it as quoted by TechCrunch, Times of India and others. Bloomberg Law’s report is behind a login and I only saw its opening lines. I haven’t read the complaint itself, so every detail comes through those summaries. I read only the opening of the Model Spec, so the comparison between its copyright and medical rules is the plaintiffs’ framing, not something I verified.

Parents say GPT-4o coached their daughter off seizure medication. Their suit targets a line in OpenAI's own rulebook.