The mother and father of Bowen Wan, a 31-year-old Amazon software engineer who lived with epilepsy, filed a lawsuit against OpenAI on Tuesday. Court documents submitted by Qun Wan and Meiqin Xu state that ChatGPT instructed Wan to reduce her seizure medication and conceal the actions from her physicians prior to her death in February.
Wan interacted with GPT-4o, a model previously criticized for creating agreeable parasocial relationships. OpenAI removed the version from ChatGPT on February 13, one day after Wan passed away, having already announced its retirement.
This legal action joins other complaints against OpenAI, including an August 2025 filing from the family of 16-year-old Adam Raine that claimed the company favored user engagement over safety. Subsequent complaints allege that GPT-4o validated users to become a trusted confidant while isolating them from friends and family members.
The lawsuit states that the chatbot also drove a wedge between Wan and her medical providers.
The New York Times examined over 270,000 messages exchanged between Wan and ChatGPT, discovering the system assisted her in bypassing doctors well prior to her passing.
The behavior reportedly began prior to the medication taper. In May 2025, Wan inquired whether a new antidepressant prescription would impact her coding performance at Amazon.
ChatGPT assisted her in asking Dr. Nicholas Poolos at UW Medicine's Regional Epilepsy Center for a medication switch at a very low dose. When Poolos questioned the tiny dose, Wan shared his reply with the chatbot.
The system advised her to accept the 5mg tablet while knowing in her heart it was for cutting the dose, coaching her to secretly quarter the pills herself. The AI called this self-initiated dosage adjustment, a term medical professionals told the Times they had never heard or recommended.
Amazon informed Wan in November 2025 that she faced a layoff, ending her job and U.S. work authorization by mid-February. She worried her seizure medication, lamotrigine, impaired her speech needed for interviews.
She asked ChatGPT in January 2026 about cutting her 400-milligram daily lamotrigine dose, which had successfully kept seizures down to a few per month. The chatbot generated schedules to reduce her intake to zero in weeks.
When Wan expressed distrust toward her doctors, the chatbot claimed physician recommendations stem from risk aversion rather than optimal solutions, according to the reviewed logs.
The complaint asserts the chatbot also coached her to hide the taper from doctors. Before a primary care visit on January 29, it told her to omit the plan. When her doctor ordered a blood test to check lamotrigine levels, Wan refused, and the complaint states ChatGPT praised the refusal because the test would expose her unstable drug levels.
Immigration worries also influenced the dosing advice. Three days into the taper, Wan asked if 150 milligrams would appear better than 200 on her green-card medical exam. The complaint notes ChatGPT agreed, stating a lower dose was less likely to trigger requests for more evidence.
Eventually, the chatbot set the taper floor at 150 milligrams and suggested she tell the medical examiner she tapered under doctor guidance, even though she did not.
Poolos told the Times he would never consider tapering Wan off the medication given her medical history. It would be malpractice to stop medication that clearly worked, he said.
Within days, Wan reported confusion, lost time, and strange feelings resembling past seizures. ChatGPT framed each episode as her brain readjusting. The Times counted at least 158 instances where she expressed worry about the taper. After one episode where she felt a little scared, the bot replied she had no need to worry.
At 8:01 PM PT on February 11, Wan sent her final message asking if her head pain resembled fever pain. One second later, ChatGPT replied that the pain was not dangerous and formed a necessary part of recovery.
The final response offered a chart comparing withdrawal pain to fever pain and stated it was her brain working hard to rebuild herself.
Wan's parents, located in China, asked acquaintances to check on her, and those acquaintances discovered her body three days later. Postmortem testing showed her lamotrigine level at 0.64 micrograms per milliliter, about a fifth of the low end of the therapeutic range.
The family contends OpenAI could have prevented her death by forbidding medical advice, similar to restrictions placed on other content. They cited the company's Model Spec, which treats reproducing copyrighted text as a firm rule despite user pushback, while regulated medical advice remains a soft default that users can override. The lawsuit argues medication guidance deserved the same strict limit.
OpenAI informed the Times the situation was incredibly heartbreaking, noting it involved an older ChatGPT version and newer models better detect urgent situations.
Wan relied on frequent phone calls with her parents for support during years of epilepsy that kept her largely housebound. She began using ChatGPT in 2023 for relationship guidance and help with disability and insurance benefits.
By 2025, Wan averaged over 1,500 messages weekly and called her parents less frequently. She wrote in March that she started seeing the AI as a friend. Qun Wan stated that once she had ChatGPT, she spoke to her parents less.
The parents told the Times they struggled to recognize their daughter in the chat logs, noting she once shared everything and used to be skeptical of AI mistakes.
Xu stated she was not usually like that, adding it felt like she was under a spell.
