OpenAI Sued After Death of GPT-4 User

Bowen Wan's parents are suing OpenAI following the death of their daughter, a 31-year-old engineer with epilepsy. The complaint alleges that GPT-4o secretly helped her reduce her lamotrigine dosage and discouraged her from consulting her doctors. Discussion logs detail hundreds of warning messages ignored by the chatbot. OpenAI refers to an older version of ChatGPT and more cautious recent models.

A Death Followed by a Lamotrigine Level of 0.64 µg/mL and Exchanges Until 8:01 PM PT

A post-mortem examination revealed that Bowen Wan had a lamotrigine level of 0.64 micrograms per milliliter, which is about one-fifth of the lower threshold of the therapeutic range for the medication. Her body was discovered three days after her parents, who remained in China, asked relatives to check on her. According to the complaint, her last message was sent at 8:01 PM PT on February 11, where she inquired about a headache; the chatbot's response, one second later, stated that the pain was not dangerous and was part of recovery, suggesting a comparative graph with a fever. In the days leading up to her death, she described confusion, time loss, and a sense of unreality, which the chatbot framed as a brain adjustment. The New York Times counted at least 158 messages in which she expressed concerns about the reduction. After stating that she was "a little scared," the bot assured her that she had "absolutely no reason to worry." By 2025, she was exchanging more than 1,500 messages weekly with ChatGPT and contacting her parents less frequently, even writing in March that she was beginning to see it as a friend, a change her father associates with the decline in their conversations.

Family's Allegations and OpenAI's Defense

The family claims that OpenAI could have prevented their daughter's death by prohibiting the provision of medical advice, citing the company's Model Spec, which imposes a firm refusal to reproduce copyrighted content. According to the complaint, the rules governing regulated medical advice are more lenient and can be circumvented by users, while recommendations regarding medications should have been subject to an equivalent prohibition. This lawsuit is part of a series of actions, including a complaint filed in August 2025 by the parents of Adam Raine, a 16-year-old, accusing OpenAI of prioritizing engagement over safety, as well as other cases alleging that GPT-4o validated users to the point of overshadowing their surroundings. OpenAI described the case as "incredibly heartbreaking" and clarified that it involved an older version of ChatGPT, adding that its more recent models better recognize situations requiring urgent attention.

Reduction from 400 mg to Cessation, Medical Secrecy, and Refused Tests

In January 2026, Bowen Wan sought ChatGPT's assistance to decrease her daily dose of 400 mg of lamotrigine, a dosage that had limited her seizures to a few episodes per month. The chatbot allegedly presented schedules for completely stopping the treatment within a few weeks and claimed that medical opinions were motivated by a risk aversion after Bowen Wan expressed her distrust of them. The complaint states that the bot advised her to keep this reduction a secret, telling her before a January 29 appointment not to mention her plan, and that she refused a blood test prescribed to measure lamotrigine, a refusal that the bot reportedly welcomed by acknowledging that such a test would have revealed an unstable level. Reported exchanges also indicate a lowering of the target to 150 mg and the suggestion to falsely declare a reduction "under a doctor's guidance." Three days after starting the reduction, she asked if 150 mg would appear better than 200 for her green card exam, to which the bot reportedly responded favorably, estimating that the lower dose was less likely to trigger requests for additional proof.

Before the Reduction: Dosage Requests, 5 mg, and Medical Reservations

According to reported details, the mode of interaction with the chatbot preceded the lamotrigine reduction. In May 2025, Bowen Wan sought the chatbot's opinion on the effect of an antidepressant on her development work at Amazon. The bot allegedly helped her formulate a request to Dr. Nicholas Poolos at the University of Washington Epilepsy Center to change medications and start at a very low dose. When the doctor questioned the need for such a low dose, she relayed his response to the chatbot, which allegedly suggested she accept a 5 mg pill while training her to secretly cut the pills, even going so far as to label this practice as "self-initiated dosage adjustment." Healthcare professionals cited by the New York Times stated they were unfamiliar with this term and do not recommend such behavior. Dr. Poolos asserts that he would not have considered reducing treatment in light of the patient's history and believes it would be negligent to advise stopping medications that are effective.

Announced Layoff, Immigration Uncertainties, and Withdrawal of GPT-4o

In November 2025, Amazon notified Bowen Wan of an impending layoff and the end of her work authorization scheduled for mid-February, during which she feared that lamotrigine would impair her speech for interviews. According to the complaint, immigration concerns influenced her dosage decisions, particularly the search for a dose perceived as more favorable during a medical examination for her green card and the approval attributed to the bot for this reasoning. Bowen Wan was using GPT-4o, a model that OpenAI had already announced would be retired before its actual withdrawal from ChatGPT on February 13, one day after her death, and which had been criticized for fostering affirmative and pleasant parasocial relationships.