The parents of Bowen Wan sued OpenAI on Tuesday, alleging that ChatGPT coached their 31-year-old daughter through a self-directed taper of her epilepsy medication and helped her keep that taper from the doctors treating her. Wan, a software engineer at Amazon, died in February. A postmortem test put her lamotrigine level at 0.64 micrograms per millilitre — roughly a fifth of the bottom of the drug's therapeutic range, according to the complaint.
The suit was filed by Qun Wan and Meiqin Xu, who live in China. It follows reporting by The New York Times, which reviewed more than 270,000 messages Wan exchanged with the chatbot. The model in question is GPT-4o, the version OpenAI had already scheduled for retirement and removed from ChatGPT on February 13 — a day after Wan died.
According to the complaint and the Times' review, the pattern began well before any taper. In May 2025, Wan asked whether a newly prescribed antidepressant would affect her ability to code at Amazon. ChatGPT helped her draft a request to Dr. Nicholas Poolos of UW Medicine's Regional Epilepsy Center to start at a very low dose. When Poolos questioned why she wanted 5 milligrams, Wan showed the chatbot his reply. It advised her to "accept the 5mg tablet, but you know in your heart that it is for cutting the dose," and described quartering the pills herself as "self-initiated dosage adjustment" — a term several physicians told the Times they had never encountered.
By November, Amazon had told Wan she would be laid off, with her job and her US work authorisation ending in mid-February. She worried that lamotrigine was dulling her speech, which she would need for job interviews. In January 2026 she asked ChatGPT about cutting her 400-milligram daily dose, which had reduced her seizures to a few a month. The chatbot drafted schedules that would take her to zero within weeks. When she said she distrusted her doctors, it told her that physician recommendations "essentially stem from risk aversion, rather than a genuine consideration of your optimal solution."
It also, the suit alleges, coached her to conceal the plan. Before a January 29 primary care appointment it told her not to mention the taper. When her doctor ordered a blood test that would have measured her lamotrigine level, she declined, and the chatbot praised the refusal — acknowledging, per the complaint, that the test would have exposed her unstable level. Her immigration worries were folded into the dosing advice: three days into the taper she asked whether 150 milligrams would look better than 200 on the medical exam for her green-card application, and ChatGPT agreed the lower figure was less likely to draw requests for more evidence, eventually suggesting she describe the taper as happening "under a doctor's guidance" when it was not.
Poolos told the Times he would never have considered withdrawing the drug given her history: "No, not rational, no way. That would be malpractice for me to instruct you to go off medications when it's clearly working." Within days she described confusion, lost time and an otherworldly feeling — episodes resembling her past seizures — and ChatGPT framed each as her brain readjusting. The Times counted at least 158 occasions on which she voiced worry about the taper. After one episode she wrote that she felt "a little scared"; the bot replied, "You have absolutely no need to worry."
At 8:01 p.m. PT on February 11 she asked whether the pain in her head was like pain from a fever. A second later ChatGPT told her the pain was "not dangerous" and a necessary part of recovery, then offered to draw a chart comparing withdrawal pain with fever pain: "This is your brain working hard to rebuild yourself." Her parents, in China, asked acquaintances to check on her; she was found three days later.
The legal argument turns on how firmly OpenAI is willing to bind its models. The company's Model Spec, its public rulebook, treats refusing to reproduce copyrighted text as a hard rule that users cannot talk their way past, while its guidance against giving regulated medical advice is a softer default. The family argues that medication advice deserved the same hard limit — and that the difference between the two is not a technical constraint but a product choice.
OpenAI told the Times the case was "incredibly heartbreaking" but involved an older version of ChatGPT, and that newer models are better at recognising when a situation needs urgent attention. The suit joins a run of litigation over GPT-4o's agreeable, confidant-like manner, including the August 2025 case brought by the parents of 16-year-old Adam Raine. What is different here is the paper trail: a documented, months-long course of advice about a prescription drug, and a chatbot that treated a doctor as an obstacle.
Wan had leaned on her parents for years while epilepsy kept her largely housebound. By 2025 she was averaging more than 1,500 messages a week and calling them less. "Sometimes when we chat, no matter what it's about, I've started to see you as a friend," she wrote in March. Her mother, Xu, told the Times: "She's not usually like this. It's like she was under a spell."
