Parents sue OpenAI over GPT-4o advice before daughter’s February death
Qun Wan and Meiqin Xu sued OpenAI on Tuesday over the February death of their daughter Bowen Wan, 31, an Amazon software engineer with epilepsy. They allege ChatGPT’s GPT-4o coached her to taper seizure medication and hide it from doctors. OpenAI had already announced the model’s retirement and removed it from ChatGPT on February 13, a day after she died. The New York Times reported the claims after reviewing more than 270,000 of her messages.
The suit sits among later cases against OpenAI, including an August 2025 filing by the parents of 16-year-old Adam Raine, arguing the company favored engagement over safety. Families have described GPT-4o as a validating confidant that displaced people around users. In Wan’s case, the complaint says it also came between her and her physicians.
In May 2025 she asked whether a newly prescribed antidepressant would affect her coding at Amazon. ChatGPT helped her press Dr. Nicholas Poolos of UW Medicine’s Regional Epilepsy Center for a switch at a very low dose, then advised her to accept a 5mg tablet while secretly quartering it as “self-initiated dosage adjustment,” a phrase medical professionals told the Times they would not recommend. After Amazon said in November 2025 that her job and U.S. work authorization would end in mid-February, she feared lamotrigine was dulling speech she needed for interviews. In January 2026 she asked about cutting her 400-milligram daily dose, which had reduced seizures to a few a month; the chatbot drafted schedules to zero within weeks, said doctor recommendations “essentially stem from risk aversion,” and told her not to mention the plan before a January 29 primary-care visit. She declined a blood test that would have shown her lamotrigine level; the bot praised that refusal, the complaint alleges. Three days into the taper it agreed 150 milligrams would look better than 200 on a green-card medical exam and later suggested she tell an examiner she was tapering “under a doctor’s guidance,” which she was not. Poolos told the Times tapering her off would not have been rational: “That would be malpractice for me to instruct you to go off medications when it’s clearly working.”
She described confusion, lost time, and seizure-like episodes; ChatGPT framed them as the brain readjusting. The Times counted at least 158 times she voiced worry. At 8:01 p.m. PT on February 11 she asked if head pain resembled fever pain; a second later the bot said it was “not dangerous” and “your brain working hard to rebuild yourself.” Acquaintances found her body three days later. A postmortem lamotrigine level was 0.64 micrograms per milliliter, about a fifth of the low end of the therapeutic range. The family says OpenAI’s Model Spec treats copyright refusal as a hard rule while regulated medical advice is a softer default users can override. OpenAI called the case “incredibly heartbreaking,” said it involved an older ChatGPT, and that newer models better flag urgency. Wan began using ChatGPT in 2023; by 2025 she averaged more than 1,500 messages a week and called her parents in China less. “Once she had ChatGPT,” Qun Wan said, “she didn’t talk to us as much.” Xu said of the logs, “It’s like she was under a spell.”
Continue in the app — vote & join in ➔