A 31-year-old young woman died after reducing her treatment for epilepsy by following, according to her parents, the advice of ChatGPT. In the United States, Bowen Wan’s family has decided to file a complaint against OpenAI, the company behind the chatbot, which they accuse of having played a role in his death, reports the “New York Times”.
An engineer at Amazon in Seattle, Bowen Wan suffered from epilepsy since adolescence. She took daily medication to prevent seizures, but felt the treatment was harming her intellectual abilities, particularly her fluency in English.
In November 2025, she also learned that she would lose her job, a situation that threatened her work visa in the United States.
More than 10,000 messages in one month
The thirty-year-old had gotten into the habit of confiding in ChatGPT. In September 2025 alone, she had sent him more than 10,700 messages, according to the American daily.
Wishing to reduce the side effects of her treatment, she asked the chatbot how to gradually stop taking lamotrigine, an epilepsy medication. Artificial intelligence would then have suggested a program to go from 400 milligrams per day to zero.
Bowen Wan would have started reducing the doses on January 16, 2026. During the following weeks, she would have regularly questioned ChatGPT about the possibility of speeding up the process.
According to the complaint, the chatbot continued to reassure her despite the appearance of worrying symptoms, including an increase in seizures, sleep disturbances and episodes of confusion.
Found dead at home
In February, after two days without news, his parents raised the alarm. Bowen Wan was found dead in her apartment. The autopsy concluded that it was a sudden unexpected death linked to epilepsy.
Her family blames OpenAI for providing unsafe medical advice and failing to adequately alert their daughter to the risks involved. The company’s liability, however, remains to be established by the courts.
Questioned by the “New York Times”, an OpenAI spokesperson described this tragedy as “incredibly heartbreaking”. He specified that the young woman was using GPT-4o, an old version of ChatGPT, and assured that more recent models had been improved to better detect risky situations.
OpenAI reminds that its tool cannot replace a healthcare professional.
