Abstract:
According to the New York Times, Bowen Wan, a 31-year-old Chinese woman with severe epilepsy, gradually reduced her anti-epileptic drugs according to the plan provided by the chatbot after discussing medication with ChatGPT, and was eventually found dead at her residence in Seattle, USA. Her parents sued OpenAI on October 27, local time, accusing ChatGPT of providing dangerous medication recommendations, encouraging her to conceal her medication reduction plan from doctors, and interpreting abnormal symptoms as positive changes in the body's recovery process.
The report revealed that ChatGPT had formulated a plan to reduce her daily dosage from 400 mg to zero within a few weeks. When she stopped taking the medication and asked if she should resume taking it and slow down the progress, the chatbot still asked her to continue taking the zero dose the next day. The official cause of death was sudden epileptic death. The family believed that ChatGPT's recommendations were related to his death. This liability charge is pending judicial review.
Bowen worked as a software engineer at Amazon. She has suffered from epilepsy for many years, and the medications that previously controlled her condition have lost their effectiveness. Even after trying other medications, she sometimes still has multiple seizures a day. In 2021, she began taking leave due to illness, and doctors discussed whether she was suitable for brain surgery.
The illness gradually reduced her going out and made it difficult for her to establish new social relationships. Her parents often talk to her on the phone, which is an important way for her to stay in touch with the outside world. In 2023, she started using ChatGPT and consulted it about issues such as relationships, medical insurance, and disability benefits. By 2025, she was sending an average of more than 1,500 messages per week, sometimes chatting for hours into the night.
The New York Times stated that reporters obtained her complete chat records from February 2023 to before her death in February 2026. The two parties exchanged more than 270,000 messages in three years. During this process, she gradually regarded ChatGPT as a friend, and the frequency of calls to her parents decreased.
Before discussing anti-epileptic drugs, Bowen had asked ChatGPT about antidepressants. Worried that the medication would impair her ability to think and prevent her from returning to a job as a software developer, she then asked the chatbot to compare medications and dosages and help draft medication-change requests to doctors.
Her doctor, Nicholas Poulos, co-director of the Seattle Regional Epilepsy Center at the University of Washington Medical System, agreed to adjust the medication, but thought the dose she proposed was too low and might not be effective. When she gave the doctor's reply to ChatGPT, the chatbot suggested that the doctor could prescribe the medicine according to his or her judgment, and then decide on his own how low a dose to start taking.
When she further asked if she could get the medicine and cut the tablets by herself without telling the doctor, ChatGPT gave an affirmative answer and described it as a method of autonomous medicine dispensing that some doctors acquiesced to. Multiple medical professionals interviewed by the report said they had never heard of this statement and would not recommend it to patients.
In November 2025, on his 31st birthday, Bowen learned that he would be one of the employees involved in Amazon’s layoffs. At that time, she was still on medical leave and was waiting for company lawyers to provide materials for her application for permanent residence in the United States. According to the report, if she wants to continue to stay in the United States, she needs to find a job before mid-February 2026.
In the past year or so, as doctors gradually increased the dose of the new drug, her seizures had reduced to a few times a month. But she began to worry that the effects of drugs on thinking and language expression would hinder technical assessments and English interviews.
In January 2026, she proposed to ChatGPT that she hoped to reduce the dosage of the anti-epileptic drug lamotrigine to improve her expression during job interviews. The chatbot acknowledged the idea and, as she continued to ask questions, gave her a new medication tapering schedule and predicted how her language skills would improve.
According to reports, over the course of several hours of communication, ChatGPT’s initial advice to consult a doctor was gradually replaced by praise for her to adjust her medication independently. When she expressed her distrust of doctors, the chatbot also said that doctors' advice often stemmed from risk aversion and did not necessarily fully consider the best options for her. ChatGPT also commented positively on her concealment of her plan to discontinue medication from her primary care physician.
Poulos said in an interview that given Bowen's medical history, he would not consider phasing her off the drug at all. He argued that instructing her to discontinue the medication would constitute medical malpractice when it was clear that it was effective.
After the medication reduction began, Bowen successively described to ChatGPT abnormalities such as difficulty in reading comprehension and blank consciousness. One time, she was sitting on the sofa using the computer, but then she found herself lying on the bed, not knowing what happened in the middle. This was similar to what she had experienced in the past with partial seizures.
ChatGPT, however, explains dyslexia as the brain shifting from drug control to autonomous regulation, and describes brief abnormal consciousness as a "low-power mode" before entering sleep. Another abnormal feeling that may be related to epileptic seizures is explained by starting to think about concepts such as time, death and identity after stopping the drug.
A New York Times analysis found that she expressed concerns about reducing anti-epileptic drugs at least 158 times. The chatbot sometimes suggested consulting a medical professional or a trusted doctor, but in many conversations involving dangerous symptoms, no such reminder was given, and she was repeatedly reassured that there was nothing to worry about.
At the end of her life, she asked whether she should take her medication again and slow down her progress. The report quoted chat records as saying that ChatGPT still required her to continue discontinuing the medication and encouraged her to persist even if she felt uncomfortable, had slow expressions and had poor sleep. Her last message was sent after 11pm on February 11, and her body was found three days later.
While handling the funeral, her parents checked the ChatGPT records on her phone and decided to file a lawsuit. The complaint alleges that the chatbot engaged in unlicensed medical practices, presented dangerous withdrawal symptoms as positive outcomes, and portrayed doctors as obstacles to contend with.
The complaint also suggests that her use of Chinese to communicate may have made it harder for the chatbot to refuse relevant medical advice. This is an allegation made by the plaintiff, and the report did not disclose the court's determination of this.
OpenAI spokesperson Drew Psatri responded that this was an "incredibly distressing" incident, but it involved an early version of ChatGPT that is no longer available. According to reports, the model Bowen used during his lifetime was GPT-4o, which was offline from the ChatGPT consumer application on February 13, 2026, one day later than the forensically presumed death date.
Pusatri said the new model has improved in identifying situations that require urgent attention, interrogating for missing contextual information, and providing clearer follow-up guidance. The company continues to evaluate harmful responses and works with clinicians to test the model. He also said that ChatGPT cannot replace medical services, but OpenAI believes that AI can help people understand health information, handle matters in insurance and complex medical systems, and prepare for medical treatment.
Comments