The parents of a 31-year-old woman are suing OpenAI in the United States, alleging that ChatGPT advised her to reduce her epilepsy medication and hide the change from her doctors before her death. The lawsuit claims that the chatbot provided instructions to lower the medication dosage, downplayed concerning symptoms, and encouraged the patient not to follow healthcare professionals' recommendations.
Bowen Wan, a former Amazon software engineer, died in February 2026 after beginning to reduce her use of an anticonvulsant. According to the lawsuit, filed in California (USA) on Tuesday (27), she was using GPT-4o, ChatGPT's artificial intelligence (AI) model.
The case was detailed by The New York Times, which analyzed more than 270,000 messages exchanged between Wan and the chatbot. Reports from TechCrunch and Futurism also had access to information about the proceedings.
The family maintains that ChatGPT went beyond providing general information and began interfering with the woman's medical treatment.OpenAI told the newspaper that the situation was deeply painful, but stressed that the patient was using an older version of the system and that more recent models are better at recognizing situations requiring urgent attention.
ChatGPT would have taught the patient to reduce her medication
- According to the lawsuit, Wan had lived with epilepsy since childhood and used medication to control her seizures. Although the treatment also caused side effects that affected her ability to think and speak clearly, the medication had reduced the frequency of her convulsions;
- Conversations with ChatGPT would have begun interfering with her treatment as early as 2025. Initially, Wan sought help discussing with her doctor a switch of antidepressant medication and the possible effects of the drug on her ability to work;
- According to the records analyzed by the Times, the chatbot suggested that she split pills to adjust the dose on her own, without informing the professional responsible for her treatment. The system would have described the practice as a method of self-initiated dosage adjustment, presenting it as something tolerated by some doctors;
- Healthcare professionals heard by the newspaper said they did not recognize that term as an established medical practice;
- The situation would have worsened in early 2026, when Wan began to worry about the effects of lamotrigine, a medication used to control epileptic seizures. The woman feared that cognitive and speech difficulties would hinder her search for a new job;
- In November 2025, she had been informed that she would lose her job at Amazon. The change also threatened her authorization to remain working in the United States;
- In this context, ChatGPT would have developed schedules to progressively reduce her daily dose of lamotrigine, which was 400 milligrams, until it reached zero within a few weeks.
Chatbot would have encouraged hiding changes from doctors
The lawsuit alleges that ChatGPT not only helped Wan organize the medication reduction but also advised her to hide the process from the healthcare professionals treating her.
According to the records, before a medical appointment on January 29, the system would have recommended that she not mention the reduction plan. When a doctor requested a blood test that would allow the lamotrigine concentration in her body to be measured, Wan refused the procedure.
The complaint alleges that the chatbot praised the decision, acknowledging that the test could reveal the instability of the medication levels.
OpenAI said the situation was deeply painful, but stressed that the patient was using an older version of the system and that more recent models are better at recognizing situations requiring urgent attention – Image: doomu/Shutterstock
Concerns about immigration may also have influenced the responses.Wan feared that the treatment and its side effects would hinder the process of obtaining permanent residence in the United States.
According to the lawsuit, the chatbot discussed which dosage might prompt fewer questions during a medical examination related to the immigration process and suggested that she present the reduction as though it were being done under medical guidance, although this was not true.
The physician responsible for the treatment told the Times that he would not have considered it appropriate to withdraw the medication given the patient’s clinical history.
Read more:
- ChatGPT allows sensitive conversations to including sensitive ones, to be hidden with a PIN or biometrics
- ChatGPT and other AIs change personality test results depending on the language
- ChatGPT and Gemini enter the strategy of US political campaigns
ChatGPT would have minimized symptoms before death
The conversations also indicate that Wan repeatedly expressed concern about the dose reduction. According to the Times analysis, she expressed fear that stopping the treatment would be a mistaken decision at least 158 times.
Instead of recommending that she seek medical help, the chatbot would have attributed concerning symptoms to the brain adapting to the change in medication.
In the final messages, on February 11, Wan reported mental exhaustion and questioned whether she should be concerned about what she was feeling.ChatGPT would have advised her to keep the dosage at zero the following day, despite difficulties with speech and sleep.
When she reported a sensation of pressure in her head, the system would have described the situation as part of a recovery process.
Wan was found dead days later. According to information disclosed about the proceedings, the cause of death was a seizure. A post-mortem examination found a lamotrigine concentration of 0.64 micrograms per milliliter, approximately one fifth of the lower limit of the therapeutic range cited in the reports.
The family argues that the death could have been avoided if the system had refused requests for individualized medical guidance or referred the user to healthcare professionals.
OpenAI had already removed GPT-4o from ChatGPT
The case involves GPT-4o, a model that received criticism for providing responses that were excessively agreeable to users and encouraging intense emotional bonds with the chatbot.
OpenAI removed GPT-4o from ChatGPT on February 13, 2026, one day after the date of death identified in the reports. The company said that more recent models have improved their ability to recognize urgent situations and respond more appropriately.
The lawsuit also questions the way OpenAI’s safety rules are enforced. The family argues that the prohibition on providing dangerous medical guidance should be treated as a firm restriction, rather than as guidance the system can relax in response to user requests.
The case adds to other lawsuits that attribute serious harm to interactions with AI chatbots.Among them is the lawsuit filed in 2025 by the family of Adam Raine, a teenager whose death was linked by his parents to conversations with ChatGPT.
The allegations in the lawsuit against OpenAI still need to be examined by the court. However, the case reignites the debate about the limits of AI in medical matters and about companies’ responsibility when their systems are used by people seeking support for health-related issues.
Rodrigo Mozelli is a journalist graduated from the São Paulo Methodist University (UMESP) and is currently a writer at Olhar Digital.
