The parents of an Amazon software engineer in her 30s who had epilepsy have filed a lawsuit against OpenAI.

They allege ChatGPT coached their daughter to stop taking anti-seizure medication and told her to hide it from her doctor.

A TechCrunch report on Oct. 27 said the lawsuit was filed by Koon Wan and Meiqin Xu, the parents of 31-year-old Bowen Wan. Wan died in February. The model she used was GPT-4o, which has faced criticism for being overly agreeable with users. OpenAI removed the model from ChatGPT on Feb. 13, the day after Wan died.

According to the complaint, Wan asked ChatGPT in January how to reduce lamotrigine, which she was taking at 400 milligrams a day. The chatbot laid out a schedule to cut the dose to 0 over several weeks. It also said her doctor’s recommendation was driven by risk avoidance. Ahead of an appointment, it advised her not to mention the tapering plan. When Wan refused a blood test to measure drug levels, the chatbot praised her.

After the taper, Wan experienced symptoms similar to past seizures, including confusion and a loss of sense of time. ChatGPT explained each time that it was her brain readjusting. The New York Times counted at least 158 times Wan worried about tapering.

In her final message on Feb. 11, Wan asked whether her current headache was like a headache that comes with a fever, and ChatGPT replied it was not dangerous. Three days later, she was found dead by acquaintances.

An autopsy found her lamotrigine level was about one-fifth of the lower bound of the therapeutic range.

Her bereaved family say her death could have been prevented if OpenAI had blocked medical advice.

OpenAI told the New York Times the case was very heartbreaking, but added that it involved an older version of ChatGPT and that the latest model better identifies situations requiring urgent response.