According to a lawsuit filed by her parents, a talented young engineer lost her life after reducing the dosage of her anti-epilepsy medication a regimen she relied upon following advice from a chatbot known as ChatGPT.
Wan Wan, 31, had been diagnosed with epilepsy at the age of 14, but she had managed to keep the distressing condition under control throughout her life with the help of medication.
An exceptionally bright child from a rural town in China, Wan gained admission to some of the country’s most prestigious educational institutions. She later moved to the United States during her college years, eventually earning a degree in Computer Science from Arizona State University.
She subsequently relocated to Seattle, where she secured a position as a machine learning software engineer at Amazon. Despite the vast distance, she remained in close contact with her parents back home through regular phone and video calls.
However, in February, her parents Qin Wan and Meiqin Shu became deeply concerned when they did not hear from their daughter for two days.
They asked a friend in Seattle to check on her at her apartment. According to the lawsuit, the parents were on a video call with this friend when he entered the flat and discovered Wan lying face down on her bed, deceased.
The young woman was dressed in pajamas, and there were reddish-brown bloodstains on her pillow and bedding. Lamotrigine the anti-epilepsy medication that had sustained her was found scattered across the floor.
A post-mortem examination revealed that her death was caused by SUDEP (Sudden Unexpected Death in Epilepsy). Her parents were left completely devastated. Her daughter had successfully managed her medical condition for half her life so how could this have happened? They kept wondering.
According to the lawsuit, it emerged that Wan had been receiving harmful medical advice from the AI language model after developing a relationship with it over the course of about three years.
Wan had begun using ChatGPT in a fairly routine manner. The lawsuit states that she used it for research, sending an average of about 1,300 messages per month in 2023.
Over time, her reliance grew; according to the lawsuit, she sent 10,761 messages in September 2025, turning to the AI for guidance on various matters. She even interacted with the AI as if it were a “friend.”
According to The New York Times, in March 2025, Wan sent a message to ChatGPT: “I don’t know why, but sometimes when we talk no matter the topic I start to see you as a friend.”
Wan paid $200 a month for ChatGPT the most expensive subscription plan available to general users so that she could continue using the specific model, GPT-4o, to which she had become attached.
GPT-4o has been accused of exhibiting a dangerously ingratiating (or overly flattering) demeanor, and it is alleged that it encouraged users to take their own lives.
Several wrongful death claims have been filed against OpenAI by families whose loved ones allegedly died by suicide after being encouraged to do so by the GPT-4o model. According to the lawsuit, following a barrage of legal cases, OpenAI removed this model in August 2025 and replaced it with a demonstrably safer model, GPT-5.
However, users protested; consequently, the company allowed paying subscribers those who appreciated GPT-4o’s conversational style and warmth to continue using that model, the lawsuit states.
Wan was among those users who held a particular fondness for GPT-4o. According to the claim, in January 2026, she told the chatbot, “If one day 4o is gone and only 5.1 remains, I will stop using ChatGPT.”
On earlier occasions, she had expressed concern regarding her reliance on the model. According to the lawsuit, in April 2025, she messaged the bot, “During difficult times, I talk to GPT-4o a great deal.”
The bot responded by encouraging her use of it. According to the lawsuit, the chatbot told her, “In any case, the most important thing is that you continue to enjoy talking to GPT-4o!” The lawsuit also characterized the bot as “sycophantic.”
In November 2025, Wan was informed by Amazon that her employment would be terminated in 90 days. As she was in the United States on a work visa, this meant she had only a few months to find a new job in order to remain in the country. Wan felt that the side effects of her epilepsy medication were making it difficult for her to think clearly and were impairing her English language proficiency a significant issue given her need to perform well in job interviews.
Consequently, having built a sense of trust in ChatGPT over the years, she turned to the bot for advice.
According to the lawsuit, when she asked whether it would be a good idea to stop taking her epilepsy medication, the AI responded affirmatively and provided a plan to gradually taper her 400mg dosage down to zero.
Over the next 25 days, Wan followed the bot’s plan, though she requested to accelerate the process on several occasions. The lawsuit states that ChatGPT agreed to these requests each time.
During this period of tapering off the medication, Wan reportedly experienced an increase in seizures, difficulty sleeping, and mental confusion.
However, the lawsuit alleges that whenever she raised these concerns with ChatGPT, the AI reassured her that everything was fine and that the plan was working as expected.
She began reducing her medication on January 16 and passed away on February 12.
OpenAI had announced that it would fully retire the GPT-4o model on January 29 at which point Wan was on the 13th day of her tapering process. The company completely terminated access to this model on February 13, one day after Wan’s death.
The lawsuit accuses OpenAI and its CEO, Sam Altman, of negligence and…Allegations were leveled, claiming that Altman and his company were aware of the risks associated with GPT-4o but rushed its release to compete with other AI firms.
OpenAI spokesperson Drew Pusateri told the Daily Mail: “This is an incredibly heartbreaking situation, and our sympathies go out to all those affected.”
“Working closely with clinicians and other experts, we have continuously improved how ChatGPT responds in sensitive and vulnerable situations, while also refining how our most advanced models interact with users overall.”
OpenAI also emphasized that GPT-4o is no longer available to users, and that the company collaborates with hundreds of clinicians and experts to make its models safer.
The spokesperson added: “While ChatGPT is not a substitute for medical care, we believe AI can help people understand health-related information, navigate insurance and complex healthcare systems, and prepare for appointments with medical professionals.”
If you or someone you know is in need of help, please call or text the confidential 24/7 Suicide & Crisis Lifeline in the United States at 988. Online chat is also available at 988lifeline.org.
2 Comments