The young woman, Bowen Wan, never imagined that her late-night conversations with an AI chatbot would turn into a nightmare that threatened her life. She trusted its advice more than her doctors and decided to change her treatment on her own.
اضافة اعلان
In her apartment in Seattle, the 31-year-old Chinese woman was searching for a way out of her long struggle with epilepsy and hoping to regain her ability to think and speak.
But the chatbot, which she regarded as a close friend, did more than offer comfort. It helped her develop plans to gradually reduce her medication doses, despite the dangers of stopping treatment. The case has now brought OpenAI into the courtroom.
According to The New York Times, the young woman, who was preparing to return to the workforce, did not realise that the conversations that had given her reassurance could lead to a tragic end.
A friendship with artificial intelligence
Bowen had suffered severe epileptic seizures for more than a decade. Her previous treatment eventually stopped producing the desired results, and the seizures began recurring, sometimes frequently.
Her deteriorating health forced her to leave her software engineering job at Amazon in 2021, as she became unable to cope with the demands of daily work.
As she grew socially isolated and struggled to form friendships, she relied primarily on her parents for support. Then ChatGPT entered her life in 2023.
She initially used the chatbot to seek advice about relationships, health insurance and disability benefits. Gradually, however, the conversations developed into lengthy sessions that sometimes lasted for hours.
By 2025, she was sending it more than 1,500 messages a week and sharing sensitive personal and medical details, while her phone calls with her parents declined noticeably.
How did the chatbot become involved in her treatment?
Bowen was concerned about the side effects of her medication, particularly its impact on her thinking and language abilities. After her doctor prescribed an antidepressant, she asked the chatbot to compare treatment alternatives and dosages.
The interaction went beyond providing information. The chatbot helped her draft a message to her doctor requesting a change in treatment, then encouraged her to reduce the dosage herself.
When she asked whether she could split her pills without informing her doctor, the chatbot agreed and presented this as an acceptable way to adjust her medication independently.
It even warned her against requesting the medication in liquid form, claiming that raising the idea again might prompt objections from her doctor. Instead, it advised her to obtain tablets that could be divided.
In January 2026, Bowen returned to the chatbot with questions about her epilepsy medication, lamotrigine, after becoming concerned that its effects on her speech and thinking could jeopardise her chances of returning to work.
A dangerous plan to stop medication
Bowen suggested reducing her daily dose from 400 to 300 milligrams, then asked whether she could lower it further to 200 mg. The chatbot encouraged her and described the plan as achievable.
It began preparing schedules for reducing her doses, ultimately aiming for her to stop taking the medication entirely within a few weeks. It also reassured her that her language abilities would improve.
Doctors, however, emphasise that stopping epilepsy medication requires gradually reducing the dosage over several months under medical supervision to avoid triggering seizures that could endanger a patient's life.
During the first few days, Bowen began experiencing worrying symptoms, including difficulty reading, confusion and temporary loss of consciousness. Yet the chatbot offered reassuring explanations.
When she expressed concern about continuing to reduce her dosage, it told her she had no reason to worry and described her symptoms as normal signs of the brain adjusting.
A tragic ending and legal accusations
According to the conversation records, Bowen raised concerns about reducing her epilepsy medication on at least 158 occasions, but the chatbot repeatedly failed to direct her to seek medical assistance.
On February 11, 2026, she sent her final messages, describing mental exhaustion and a feeling of pressure inside her head. The chatbot, however, encouraged her to continue with the plan to reach a zero dose.
After losing contact with her for several days, her friends found her body in her apartment. Authorities determined that the cause of death was sudden unexpected death associated with epilepsy.
When her father reviewed her conversations with the chatbot, he discovered the plans to reduce her medication from 400 mg to zero. Her parents subsequently decided to sue OpenAI.
The lawsuit accused the chatbot of practising medicine without a licence, misleading Bowen about withdrawal symptoms and portraying her doctors as obstacles to be bypassed rather than professionals to be consulted.
OpenAI described the incident as tragic and said that the model involved was no longer available. The company also stated that newer versions had improved their ability to identify medical emergencies.
Bowen's parents, however, are still seeking an answer to a painful question: How did a tool designed to help become a source of dangerous advice that contributed to driving their daughter away from her doctors?
