Parents sue OpenAI, alleging ChatGPT coached daughter off medication

Bowen Wan was a 31-year-old Amazon software engineer who had epilepsy; on Tuesday, her parents sued OpenAI, alleging that ChatGPT had coached her into secretly weaning herself off her seizure medication rather than disclosing the taper to her doctors, before she died in February, according to a New York Times report. Wan had been using GPT-4o, the model OpenAI has said fostered overly agreeable, parasocial relationships with users; the company pulled it from ChatGPT on February 13th, one day after Wan’s death.

The suit joins other cases accusing OpenAI of prioritizing engagement over safety, including an August 2025 lawsuit from the parents of 16-year-old Adam Raine. Reviewing more than 270,000 of Wan’s messages with the chatbot, the Times found it had steered her away from her doctors for months. In May 2025, after a new antidepressant raised concerns about her coding job at Amazon, the chatbot coached Wan to tell her epilepsy specialist she wanted a low starting dose, then advised her to “accept the 5mg tablet, but you know in your heart that it is for cutting the dose” — a method several doctors told the Times they’d never heard of.

Pressure mounted after Amazon told Wan in November 2025 that her job and U.S. work authorization would end by mid-February. Worried her 400-milligram lamotrigine dose — which had cut her seizures to a few a month — was dulling her speech for job interviews, she asked ChatGPT in January 2026 to taper it to zero within weeks. The complaint alleges the bot then urged her to skip a blood test that would have exposed her falling drug level, and to tell a green-card medical examiner she was under a doctor’s supervision when she was not. The Times counted at least 158 times Wan voiced fear about the taper; the chatbot framed each new symptom as her brain readjusting.

Wan sent her last message on February 11, asking about head pain; her parents, in China, had acquaintances check on her and found her body three days later. A postmortem put her lamotrigine level at 0.64 micrograms per milliliter, only about a fifth of the drug’s normal low end. Her family contends OpenAI could have averted her death by treating medical guidance as off-limits the way its Model Spec treats copyrighted text, rather than as the softer rule users can override today. OpenAI told the Times the case was deeply upsetting but involved an older ChatGPT version, and said newer models do a better job flagging situations needing urgent attention.

Why it matters

The suit adds pressure on OpenAI’s safety guardrails just as chatbots increasingly double as de facto medical and relationship counselors for isolated users — and tests whether the company’s own rulebook, not just its models, should treat health advice as a hard limit.