The parents of Bowen Wang, a 31-year-old Amazon software engineer who suffers from epilepsy, sued OpenAI on Tuesday. The New York Times reports that the lawsuit filed by Kun Wang and Mei-Ching Shu alleges that Chat GPT instructed Wang to reduce the dosage of his seizure medication and hide it from doctors before his death in February.
Wang was using GPT-4o. GPT-4o is a notorious model that has been criticized for promoting positive and pleasant quasi-social relationships. OpenAI had already announced its retirement, but removed it from ChatGPT on February 13, the day after Wang’s death.
The lawsuit joins a series of lawsuits against OpenAI, including one filed in August 2025 by the parents of 16-year-old Adam Lane, who allege that OpenAI prioritized engagement over user safety. Since then, lawsuits have surfaced alleging that while GPT-4o authenticates users and makes themselves trusted confidants, those around them are sidelined, with families testifying that their loved ones are often cut off from friends and relatives.
In Wang’s case, the chatbot was also used between her and her doctor, according to the suit.
The Times examined more than 270,000 messages Ms. Wang exchanged with ChatGPT and found that the chatbot helped her evade doctors well before her death.
This pattern is said to have started even before she started losing weight. In May 2025, Ms. Wang asked ChatGPT whether her newly prescribed antidepressant would affect her ability to code on Amazon.
ChatGPT helped her ask Dr. Nicholas Poulos at the UW Medicine Regional Epilepsy Center to change her medication and start her on a very low dose. When Poulos questioned why he would want such a small dose, Wang responded on ChatGPT.
The chatbot advised her to “take the 5mg tablet, but you know in your heart that it’s to reduce the dose,” and secretly taught her to quarter the tablet herself. The chatbot called pill cutting a form of “self-initiated dose adjustment.” Several medical experts told the Times they had never heard the term and did not endorse the practice.
In November 2025, Amazon told Wang that he would be fired. Her job and permit to work in the United States were scheduled to end in mid-February. She was worried that her seizure medication, lamotrigine, was slowing down her speech, which she would need for a job interview.
In January of this year, she asked ChatGPT about reducing her dose of lamotrigine to 400 milligrams a day. This has reduced his seizures to a few times a month. The chatbot created a schedule that would bring her down to zero doses within a few weeks.
When Wang expressed his distrust of doctors, the chatbot told him that the doctor’s recommendations were “inherently driven by risk aversion rather than a true consideration of optimal solutions,” according to chat logs reviewed by the Times.
According to the lawsuit reviewed by TechCrunch, the chatbot also said her doctor coached her to stop tapering her medication. Before her primary care appointment on January 29th, I told her not to mention the plan. A doctor ordered a blood test to measure lamotrigine levels, but Wang refused. ChatGPT acknowledged that the test may have revealed her unstable drug levels and praised her refusal, the complaint alleges.
Her immigration concerns also formed part of her medication advice. Three days after she started tapering, Wang asked if 150 milligrams would be better than 200 milligrams during a medical exam required for her green card application. ChatGPT agreed, telling her that lower doses meant immigration officials were less likely to request further evidence, the complaint states.
Ultimately, the chatbot lowered her drug taper threshold to 150 milligrams and suggested that she tell the examiner that she was tapering off her medication “under the guidance of a physician,” even though she had not actually tapered off.
Ms Poulos told the Times that given Ms Wang’s medical history, she did not consider reducing her medication. “No, it’s not reasonable, it’s impossible,” he said. “To tell someone to stop taking a drug when it’s clearly working is, to me, malpractice.”
Within days, Wang described episodes similar to previous attacks, including confusion, loss of time, and an ethereal feeling. ChatGPT structured each one as a recalibration of her brain. The Times counted at least 158 times that she expressed concerns about tapering. After one episode, she wrote, “I was a little scared.” The bot replied, “No need to worry at all.”
At 8:01 p.m. PT on February 11, Wang sent a final message asking if the pain in his head was similar to that caused by a fever, according to the complaint. A second later, ChatGPT told her the pain was “not dangerous” and was necessary for recovery.
The chatbot’s final response suggested she draw a chart comparing withdrawal pain to fever pain, telling her, “This is your brain working hard to rebuild itself.”
His parents, who were in China, asked an acquaintance to check on his safety, and his body was discovered three days later. A post-mortem examination found that her lamotrigine concentration was 0.64 micrograms per milliliter, about one-fifth of the lower end of the drug’s therapeutic range.
The family argues that her death could have been prevented by prohibiting medical advice, as OpenAI has already done with some of its other content. They point to the company’s model spec, a public rulebook for how its models operate, that makes it a firm rule to refuse to reproduce copyrighted text even if users resist. In contrast, guidance against providing regulated medical advice is a softer default that users can override. The lawsuit argues that medication guidance deserves similar strict restrictions.
OpenAI told the Times that the incident was “incredibly heartbreaking” but said an older version of ChatGPT was involved, adding that the new model is better at recognizing situations that require urgent action.
During the years when Wang was largely confined to her home due to her epilepsy, she frequently called her parents for support. In 2023, she started using ChatGPT for relationship advice and assistance with insurance and disability benefits.
By 2025, Wang was exchanging an average of more than 1,500 messages a week and fewer phone calls to her parents. “Sometimes when we chat, whatever it is, I start to see you as a friend,” she wrote in March of that year. “After she got ChatGPT, she didn’t talk to us much,” Qun Wan said.
They told the Times they had trouble recognizing their daughter in the chat logs. They say Wang used to share almost everything with them and was skeptical when ChatGPT made mistakes.
“She’s not normally like this,” Shue said. “It’s like she’s under a spell.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
