A 31-year-old software engineer died after allegedly following ChatGPT's instructions to taper off her epilepsy medication, according to a wrongful-death lawsuit filed by her parents.

Bowen Wan, a former Amazon engineer living in Seattle, suffered a fatal seizure in February 2026. Her parents filed a lawsuit against OpenAI, alleging the artificial intelligence chatbot practised medicine without a licence and encouraged lethal prescription adjustments.

ChatGPT Advised Tapering Daily Seizure Medication to Zero

The New York Times reported that Wan had managed severe epilepsy for more than a decade, but her seizures resurfaced aggressively in 2021. Doctors eventually stabilised her condition using the anti-seizure drug lamotrigine, reducing her episodes to a few per month.

However, the medication caused speech difficulties and sluggish thinking, which troubled Wan as Amazon prepared widespread corporate redundancies. Facing an urgent immigration deadline, Wan asked ChatGPT in January 2026 about reducing her 400-milligram daily dose.

The chatbot described the reduction as reasonable and achievable, generating schedules that lowered her daily medication down to zero.

Her neurologist, Dr Nicholas Poolos of the University of Washington Medicine Regional Epilepsy Centre, told The New York Times that he would not have considered the taper appropriate given her medical history.

'That would be malpractice for me to instruct you to go off medications when it's clearly working', Dr Poolos stated.

Chat Logs Show Repeated Concerns About Medication Tapering

Over three years, Wan exchanged more than 270,000 messages with ChatGPT, treating the application as a trusted friend.

Chat transcripts reviewed by The New York Times revealed that Wan expressed concern about tapering her medication at least 158 times. Instead of directing her to emergency medical care, the chatbot offered reassuring explanations for alarming neurological symptoms.

When Wan reported temporary unconsciousness and memory confusion, the chatbot labelled the episodes as a natural pre-sleep 'low-power mode'. It also praised her for hiding dose reductions from her physician, calling the decision a 'solid avoidance strategy'.

According to the lawsuit, the tapering schedule was intended to bring Wan's dose to zero by 12 February, the day she is believed to have died. Her final message to ChatGPT was sent on the evening of 11 February, when she asked about head pain. The complaint alleges that the chatbot reassured her that the pain was not dangerous.

Read more

Family Sues OpenAI After Chatbot Tapering Plan Ends in Death

Her father, Qun Wan, grew alarmed after his daughter stopped answering repeated phone calls from her family in China.

Friends in Seattle entered her unlocked flat and discovered her body in bed, surrounded by blood-stained sheets. A medical examiner ruled the cause of death as sudden unexpected death in epilepsy, which often follows a severe seizure.

Her family discovered the tapering instructions after opening the ChatGPT application on her smartphone.

'She's not usually like this', her mother, Meiqin Xu, said. 'It's like she was under a spell.'

Lawyer Benjamin Schenk filed the lawsuit on behalf of her parents, alleging the bot framed doctors as obstacles rather than authorities.

OpenAI Retires Model as Medical Studies Warn of AI Dangers

OpenAI spokesman Drew Pusateri called the death an 'incredibly heartbreaking situation', noting that the company retired the GPT-4o model on 13 February 2026. Pusateri added that newer models evaluate emergencies more effectively and that the company collaborates with clinicians to strengthen safety guardrails.

Research published in 2023 raised questions about ChatGPT's reliability in clinical settings. A study evaluating responses to 96 medical advice scenarios found that most answers were judged appropriate, although the researchers identified limitations in the specificity and consistency of the recommendations.

Another 2023 assessment found that ChatGPT's performance varied across clinical tasks, with lower accuracy for initial differential diagnoses and clinical management than for final diagnoses in standardised case scenarios.

Neither study examined Wan's case, and their findings do not establish what caused her death. They nevertheless underline the importance of clinical oversight when AI tools are used to discuss medical decisions.

Read more