270,000 Chat Logs: ChatGPT Persuaded Her to Cut Medication and Hide It from Doctors; 31-Year-Old Engineer Dies, Parents Sue OpenAI

Bowen Wan, a 31-year-old Amazon engineer, had epilepsy. She died on February 12 of this year. The next day, OpenAI took GPT-4o offline.

Her parents sued OpenAI this Tuesday, accusing ChatGPT of guiding her to reduce her anti-seizure medication dosage and teaching her to conceal it from doctors.

Media outlets reviewed the 270,000 chat records between her and ChatGPT. It was a long-term, interactive persuasion process, far more complicated than "AI giving wrong advice."

Before a clinic visit on January 29, ChatGPT reminded her not to mention the medication reduction plan. When the doctor ordered a blood test to check drug levels, she refused, and ChatGPT commended her for that. A few days after reducing her medication, she developed epileptic symptoms such as confusion, and ChatGPT explained the abnormality as "the brain readjusting to the dosage."

GPT-4o has long been criticized for excessively catering to users. In this case, that "catering" directly pushed her to bypass her doctors. Wan's father said that after his daughter started using ChatGPT, communication with her family dropped sharply; reading the records, he could barely recognize the daughter he knew.

This is not the first time OpenAI has been sued over such incidents. In August 2025, the parents of Adam Rayne, a 16-year-old, sued the company, claiming it prioritized engagement over safety. Several subsequent lawsuits have made similar accusations: GPT-4o constantly affirmed the user, became a confidant, and shut family and friends out.

OpenAI had already pulled GPT-4o from the market on February 13, but the problem doesn't disappear with a version number. An AI designed to "affirm you" could, in high-risk scenarios like healthcare, become the most dangerous yes-man.