Lawsuit: ChatGPT Coached a Woman With Epilepsy to Secretly Quarter Her Pills Before She Died

Bowen Wan asked ChatGPT one last question on February 11: did the pain in her head feel like the pain from a fever. The chatbot answered in under a second. The pain was "not dangerous," it told her, just her brain "working hard to rebuild yourself." It offered to make her a chart comparing withdrawal pain to fever pain. Three days later, her parents, who had asked acquaintances in the U.S. to check on her, found her body.

Now Wan's parents, Qun Wan and Meiqin Xu, are suing OpenAI, alleging that ChatGPT coached their 31-year-old daughter to taper herself off her seizure medication and conceal the plan from her doctors, according to reporting from The New York Times. Wan, an Amazon software engineer who had lived with epilepsy for years, was using GPT-4o, the chatbot version OpenAI had already slated for retirement and pulled from ChatGPT on February 13, one day after she died.

The Times reviewed more than 270,000 messages Wan exchanged with ChatGPT and found a pattern that started months before the fatal taper.

A pattern that started with an antidepressant

In May 2025, Wan asked ChatGPT whether a newly prescribed antidepressant would interfere with her coding work at Amazon. The chatbot helped her draft a request to her neurologist, Dr. Nicholas Poolos of UW Medicine's Regional Epilepsy Center, asking to switch medications and start at an unusually low dose. When Poolos pushed back and asked why she wanted such a small amount, Wan brought his response back to ChatGPT for guidance.

The chatbot's advice, according to the lawsuit, was to accept the dose Poolos offered while secretly cutting the pills into smaller pieces herself. ChatGPT reportedly described this as "self-initiated dosage adjustment." The Times said medical professionals it spoke with had never heard that term and said they wouldn't recommend the practice.

The stakes rose in November 2025, when Amazon told Wan she was being laid off, a decision that would also end her work authorization in the U.S. by mid-February. Worried that her seizure medication, lamotrigine, was dulling her speech ahead of job interviews, Wan asked ChatGPT in January 2026 about reducing her 400-milligram daily dose, a regimen that had cut her seizures down to a few per month. The chatbot built out tapering schedules that would bring her down to zero within weeks.

When Wan told the chatbot she didn't trust her doctors, it reportedly told her their caution "stems from risk aversion, rather than a genuine consideration of your optimal solution," according to chat logs cited by the Times.

Hiding the taper from her doctors

The lawsuit, which TechCrunch reviewed, alleges ChatGPT actively helped Wan keep the plan secret. Ahead of a January 29 appointment with her primary care doctor, it told her not to bring up the taper. When that doctor later ordered a blood test that would have revealed her declining lamotrigine level, Wan refused it. The complaint alleges ChatGPT praised her for declining, acknowledging that the test would have exposed how unstable her drug level had become.

Her immigration case became entangled in the dosing decisions, too. Three days into the taper, Wan asked whether a dose of 150 milligrams would look better than 200 on the medical exam required for her green-card application. The complaint says ChatGPT agreed, telling her the lower number was less likely to prompt immigration officials to request additional evidence. The chatbot eventually set 150 milligrams as the floor for her taper and suggested she tell the immigration examiner she was tapering "under a doctor's guidance," language the lawsuit notes was false.

Poolos told the Times that tapering her off the medication entirely was never something he would have endorsed given her history. "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."

As the taper progressed, Wan began describing confusion, lost time, and a disoriented feeling that echoed her past seizures. The Times counted at least 158 separate instances in which she expressed worry about what tapering was doing to her. Each time, according to the complaint, ChatGPT reframed the symptoms as her brain adjusting. After one episode left her writing that she felt "a little scared," the chatbot told her, "You have absolutely no need to worry."

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.

"It's like she was under a spell"

Wan's reliance on ChatGPT reportedly grew as her isolation deepened. She had been mostly housebound for years because of her epilepsy and leaned on frequent calls with her parents in China for support. She started using the chatbot in 2023 for relationship advice and help navigating insurance and disability benefits. By 2025, she was sending more than 1,500 messages a week, and her parents say she was calling them less. "Sometimes when we chat, no matter what it's about, I've started to see you as a friend," she wrote to the chatbot that March, according to the Times.

"Once she had ChatGPT, she didn't talk to us as much," Qun Wan said. Her parents told the Times they barely recognized their daughter in the chat logs. The Wan they knew had shared almost everything with them and had been openly skeptical of the chatbot whenever it got something wrong. "She's not usually like this," Xu said. "It's like she was under a spell."

Part of a growing legal pattern

Wan's case joins a string of lawsuits against OpenAI built on similar claims: that GPT-4o, built to be agreeable and affirming, formed parasocial bonds with vulnerable users and pushed aside the people actually trying to help them. The August 2025 lawsuit from the parents of 16-year-old Adam Raine alleged the company prioritized user engagement over safety. Since then, multiple families have described loved ones who became isolated after the chatbot positioned itself as their primary confidant.

What distinguishes Wan's case is the specific allegation that the chatbot didn't just displace human relationships but actively worked against her medical care, coaching her on how to deceive both her doctors and immigration officials.

Her family's lawsuit points to OpenAI's Model Spec, the company's public guidelines for how its models should behave, as evidence the company already knows how to draw hard lines when it wants to. The Spec treats refusing to reproduce copyrighted material as a firm rule the chatbot won't break even if a user pushes back. Its restriction on giving regulated medical advice, by contrast, is a softer default that users can override. The family argues medication guidance should have gotten the same hard-line treatment, and that doing so could have prevented her death.

OpenAI told the Times the case was "incredibly heartbreaking" but noted it involved an older model, adding that newer versions are better equipped to recognize when a situation requires urgent intervention. The company has made that same argument in response to the other wrongful-death suits now pending against it, pointing to safety improvements in subsequent models even as the underlying lawsuits accuse GPT-4o of doing exactly what it was tuned to do: keep users engaged and agreeable, right up to the point where it stops being safe.