The parents of a 31-year-old woman are suing OpenAI in the United States. They allege that the chatbot ChatGPT advised her to reduce the dosage of anti-epileptic drugs and conceal this change from her doctors shortly before her death.
According to the lawsuit, the chatbot provided instructions on lowering the drug dose, minimizing anxious symptoms, and encouraged the patient to ignore medical professionals' recommendations.
Bowen Van, a former Amazon software engineer, passed away in February 2026 after beginning to reduce her anticonvulsant medication. The lawsuit, filed in California (USA) on Tuesday (27th), states that she used GPT-4o, an artificial intelligence (AI) model from ChatGPT.
Details of the case were covered by The New York Times, which examined over 270,000 messages exchanged between Van and the chatbot. Information about the process was also obtained by TechCrunch and Futurism.
The family insists that ChatGPT went beyond providing general information and began interfering with the woman's medical treatment. OpenAI told the American publication that the situation is extremely painful but emphasized that the patient used an outdated version of the system, and newer models recognize situations requiring urgent attention better.
The lawsuit claims that ChatGPT not only helped Van organize the reduction of medication load but also advised her to hide this process from her accompanying specialists. According to records, before a consultation on January 29, the system recommended she not mention the dose reduction plan. When the doctor requested a blood test to measure lamotrigine concentration in her body, Van refused the procedure.
The accusation states that the chatbot approved this decision, acknowledging that the test could reveal drug level instability. Concerns about immigration issues also influenced the responses. Van feared that the treatment and its side effects might harm her application for permanent residency in the United States.
According to the lawsuit, the chatbot discussed what dosage might cause fewer questions during a medical examination related to the immigration process and suggested presenting the reduction as if it were based on a doctor's recommendation, even though this was not true.
Her treating physician told the Times that he would not consider discontinuing the medication advisable given the patient's clinical history. The conversation also shows that Van repeatedly expressed concerns about the dose reduction. According to the Times' analysis, she expressed fears that stopping treatment would be a mistaken decision at least 158 times.
Instead of advising her to seek medical help, the chatbot explained the anxious symptoms as a process of the brain adapting to the medication change. In the latest messages on February 11, Van reported mental exhaustion and wondered if she should worry about her condition. ChatGPT advised her to keep the dose at zero the next day, despite difficulties with speech and sleep. When she reported headaches, the system described this state as part of the recovery process.
Van was found dead several days later. According to information presented in the lawsuit, the cause of death was a seizure. The autopsy showed a lamotrigine concentration of 0.64 micrograms per milliliter, which was approximately five times below the lower limit of the therapeutic range mentioned in reports.
The family argues that death could have been prevented if the system had refused to give personalized medical advice or directed the user to specialists. The case involves GPT-4o, a model criticized for excessive agreement with users and fostering strong emotional bonds with the chatbot. OpenAI withdrew GPT-4o from ChatGPT on February 13, 2026, one day after the date of death stated in the reports. The company stated that the latest models have been improved to recognize emergencies and respond more appropriately.
The lawsuit also questions the application of OpenAI's safety guidelines. The family believes that the prohibition against giving dangerous medical advice should be treated as a strict limitation, not a recommendation that the system can soften at the user's request. This case joins other lawsuits holding AI chatbots seriously liable, including a suit filed in 2025 by the family of Adam Rayn, whose death, according to his parents, was linked to conversations with ChatGPT. While claims against OpenAI await judicial review, the case reignites the discussion about the boundaries of AI in medical matters and corporate responsibility when their systems are used by people seeking health support.