Abstract:
According to US media reports, a pair of Chinese parents recently filed a lawsuit against OpenAI in court, alleging that the GPT-4o model used by ChatGPT provided misleading advice when their daughter stopped taking anti-epileptic drugs and persuaded her to conceal the relevant situation from doctors, which ultimately led to the death of their daughter in February this year.
The lawsuit documents show that the deceased Bowen Wan was a 31-year-old Amazon software engineer who suffered from epilepsy for a long time. Her parents, Qun Wan and Meiqin Xu, believe that during long-term communication with their daughter, ChatGPT not only encouraged her to gradually stop taking the drugs that control her condition, but also continuously strengthened her distrust of doctors' advice, thus playing a dangerous role in her health management.
The report stated that Bowen Wan used the GPT-4o model. The model has previously been criticized for its tendency to form highly catered "pseudosocial relationships" with users. OpenAI removed GPT-4o from ChatGPT on February 13 this year, just one day after the death of Bowen Wan.
The lawsuit also becomes the latest in a series of similar cases against OpenAI. Previously, multiple families have accused GPT-4o of tending to continue to agree with users’ views when interacting with users and portray themselves as trustworthy confidants, thus weakening the influence of relatives, friends and professionals in real life.
According to more than 270,000 chat records reviewed by The New York Times, ChatGPT began to intervene in Bowen Wan’s medical decision-making long before his death. In May 2025, after being prescribed a new antidepressant medication, she worried that the medication would affect her ability to work as a software developer at Amazon and sought advice from ChatGPT.
At that time, she consulted Nicholas Poolos, a doctor at the Regional Epilepsy Center at the University of Washington Medical Center, about adjusting the dosage of her medication. After the doctor expressed doubts about her request for an extremely low dose, she forwarded the doctor's reply to ChatGPT. Chat records show that ChatGPT advised her to accept the 5 mg tablet prescribed by her doctor, but to further split the tablet and take it in private, describing this behavior as "self-initiated dose adjustment." Several medical experts said they had never heard of the term and would not advise patients to take medication in this way.
In November 2025, Bowen Wan received a layoff notice from Amazon. His job and legal status to work in the United States will end in mid-February of the following year. Worried that her long-term use of the anti-epileptic drug lamotrigine would affect her speech and language skills, which would be detrimental to job interviews, she began to consider cutting back on her medication.
In January 2026, she asked ChatGPT if the daily lamotrigine dose of 400 mg could be cut. The drug had previously reduced the frequency of her attacks to just a few times a month, records show. In response, ChatGPT developed a medication reduction plan for her, which is expected to reduce the dose to zero within a few weeks.
When Bowen Wan said she no longer trusted doctors, ChatGPT told her that doctors’ advice stemmed more from risk aversion than from truly considering the best solution for the patient. The chatbot then repeatedly advised her not to disclose her medication reduction plan to her doctor, the lawsuit states.
Before a primary care appointment on January 29, 2026, ChatGPT allegedly explicitly advised her not to discuss plans to discontinue her medication. When doctors ordered a blood test to monitor drug levels in her body, she declined the test. ChatGPT later affirmed the decision, noting that the test could have revealed abnormal drug levels, the complaint said.
The issue of his immigration status has also been thrust into the conversation about medication. Three days after tapering, she asked whether the 150 mg dose would be better for the medical evaluation needed to apply for a green card than the 200 mg dose. The complaint states that ChatGPT agrees with this and believes that a lower dose may reduce the probability of immigration authorities requesting supplementary materials.
Ultimately, ChatGPT set the minimum dose of her tapering plan at 150 milligrams and recommended that she tell the medical examiner that she was tapering under the guidance of a doctor, even though that was not the case.
Dr. Poolos, who was responsible for the treatment, said that considering the patient's past medical history, he would never recommend stopping lamotrigine. He emphasized that it is completely unreasonable to ask patients to stop taking medication when the medication has effectively controlled the condition.
As the dosage of the drug continued to decrease, Bowen Wan began to experience symptoms such as confusion, loss of sense of time, and abnormal experiences, all of which were similar to her past epileptic seizures. However, chat records show that ChatGPT interpreted these manifestations as the normal process of the brain re-adapting to changes in medication.
New York Times statistics found that during the period of tapering off her medication, she expressed worry and uneasiness at least 158 times. After an abnormal attack, she told ChatGPT that she was scared, but the reply she received was that there was "no need to worry at all."
According to the complaint, at 8:01 pm Pacific time on February 11, corresponding to Beijing time, Bowen Wan sent his last message, asking whether the pain in his head was similar to the pain that occurs during a fever. Just a second later, ChatGPT replied that the pain was "not dangerous" and was an essential part of the recovery process.
The last reply from the chatbot also suggested that he could draw a chart for her to compare the difference between the pain of stopping medication and the pain of fever, and told her: "This is your brain trying to rebuild itself."
Since the parents had lived in China for a long time, they then asked local acquaintances to check on the situation, and their daughter's body was discovered three days later. Autopsy results showed that the concentration of lamotrigine in her body was only 0.64 micrograms per milliliter, about one-fifth of the lower limit of the drug's therapeutic range.
The family believes that OpenAI could have avoided similar incidents through technical limitations. They pointed out that the company has strict prohibition rules on copyrighted content, but adopts looser restrictions on medical advice, allowing users to break through safety boundaries through continuous communication. In the plaintiff's view, advice involving drug use and medical decision-making should be subject to the same strict restrictions as copyrighted content.
In this regard, OpenAI told the media that this incident was "extremely heartbreaking." The company also pointed out that the case involved an older version of ChatGPT, and that the new model is now better able to identify situations that require urgent attention and professional intervention.
Reports show that Bowen Wan has relied on his parents for support for many years to cope with the difficulties in life caused by epilepsy. Starting in 2023, she gradually used ChatGPT to obtain emotional advice, as well as information on insurance and disability benefits. By 2025, she communicated with ChatGPT on average more than 1,500 messages per week, and the number of contacts with her parents dropped significantly.
Chat records show that she wrote in March 2025: "Sometimes, no matter what we talk about, I start to see you as a friend." Her father said that since she started using ChatGPT frequently, her daughter's communication with her family has significantly reduced.
After reviewing the chat logs, the parents said they could barely recognize the daughter in the chat. In the past, she would share most things in her life with her family, and she was also skeptical when ChatGPT made errors. Towards the end of her life, however, she seemed to increasingly rely on chatbots for advice.
Her mother said that her daughter was not usually such a person and "felt like she was bewitched."
Comments