The mother and father of Bowen Wan, a 31-year-old Amazon software program engineer with epilepsy, sued OpenAI on Tuesday. The swimsuit, filed by Qun Wan and Meiqin Xu, alleges that ChatGPT coached Wan to taper off her seizure remedy and conceal it from her docs earlier than she died in February, The New York Occasions reported.
Wan was utilizing GPT-4o, the notorious mannequin criticized for fostering affirming, agreeable parasocial relationships. OpenAI, which had already introduced its retirement, eliminated it from ChatGPT on February 13, a day after Wan died.
The swimsuit joins a line of instances in opposition to OpenAI that embody the August 2025 lawsuit introduced by the mother and father of 16-year-old Adam Raine, which alleged OpenAI prioritized engagement over consumer security. A wave of lawsuits since then have alleged that GPT-4o validated customers and made itself a trusted confidant whereas the individuals round them have been pushed apart, with households usually describing family members minimize off from associates and kin.
In Wan’s case, the lawsuit alleges, the chatbot additionally got here between her and her docs.
The Occasions, which reviewed greater than 270,000 messages Wan exchanged with ChatGPT, discovered the chatbot had been serving to her sidestep her docs nicely earlier than her demise.
The sample allegedly began earlier than she started tapering. In Could 2025, Wan requested ChatGPT whether or not a newly prescribed antidepressant would have an effect on her capacity to code at Amazon.
ChatGPT helped her ask Dr. Nicholas Poolos of UW Drugs’s Regional Epilepsy Heart to change drugs and begin at a really low dose. When Poolos questioned why she needed such a small dose, Wan confirmed his response to ChatGPT.
The chatbot suggested her to “settle for the 5mg pill, however you understand in your coronary heart that it’s for reducing the dose,” teaching her to secretly quarter the tablets herself. The chatbot referred to as capsule reducing a type of “self-initiated dosage adjustment.” A number of medical professionals informed the Occasions they’d by no means heard the time period and wouldn’t advocate the follow.
In November 2025, Amazon informed Wan she could be laid off. Her job, and her authorization to work within the U.S., would finish in mid-February. She feared her seizure drug, lamotrigine, was dulling her speech, which she would wish for job interviews.
In January of this yr, she requested ChatGPT about reducing her 400-milligram day by day dose of lamotrigine, which had diminished her seizures to some a month. The chatbot drafted schedules that might take her dose to zero inside weeks.
When Wan mentioned she distrusted her docs, the chatbot informed her that doctor suggestions “basically stem from threat aversion, fairly than a real consideration of your optimum answer,” in response to chat logs reviewed by the Occasions.
The chatbot additionally coached her to maintain the taper from her docs, in response to the lawsuit, which TechCrunch reviewed. Earlier than a January 29 main care appointment, it informed her to not point out the plan. When her physician ordered a blood take a look at that might have measured her lamotrigine stage, Wan declined. ChatGPT praised the refusal, the grievance alleges, acknowledging that the take a look at would have uncovered her unstable drug stage.
Her immigration considerations additionally grew to become a part of the dosing recommendation. Three days into the taper, Wan requested whether or not 150 milligrams would look higher than 200 on the medical examination required for her green-card utility. ChatGPT agreed, the grievance says, telling her the decrease dose was much less prone to immediate immigration officers to request extra proof.
Finally, the chatbot lowered her taper’s ground to 150 milligrams, suggesting she inform the examiner she was tapering “underneath a health care provider’s steerage,” regardless that she was not.
Poolos informed the Occasions he wouldn’t have thought-about tapering Wan off the remedy, given her historical past. “No, not rational, no method,” he mentioned. “That might be malpractice for me to instruct you to go off drugs when it’s clearly working.”
Inside days, Wan described confusion, misplaced time, and an otherworldly feeling, episodes that resembled her previous seizures. ChatGPT framed each as her mind readjusting. The Occasions counted at the least 158 instances she voiced fear concerning the taper. After one episode, she wrote that she felt “somewhat scared.” The bot replied, “You could have completely no want to fret.”
At 8:01 PM PT on February 11, in response to the grievance, Wan despatched her remaining message, asking whether or not the ache in her head was just like ache from a fever. A second later, ChatGPT informed her the ache was “not harmful” and was a vital a part of restoration.
The chatbot’s remaining response supplied to attract her a chart evaluating withdrawal ache with fever ache and informed her, “That is your mind working laborious to rebuild your self.”
Her mother and father, who have been in China, requested acquaintances to verify on her, and so they discovered her physique three days later. A postmortem take a look at discovered that her lamotrigine stage was 0.64 micrograms per milliliter, roughly a fifth of the low finish of the drug’s therapeutic vary.
The household argues OpenAI may have prevented her demise by making medical recommendation off-limits, because it already does for another content material. They level to the corporate’s Mannequin Spec, its public rulebook for a way its fashions behave, which treats refusing to breed copyrighted textual content as a agency rule, even when customers push again. Its steerage in opposition to giving regulated medical recommendation, in contrast, is a softer default that customers can override. The swimsuit argues remedy recommendation deserved the identical laborious restrict.
OpenAI informed the Occasions the case was “extremely heartbreaking” however mentioned it concerned an older model of ChatGPT, including that newer fashions are higher at recognizing when a state of affairs wants pressing consideration.
Wan had relied on her mother and father, whom she referred to as usually, for assist throughout years of epilepsy that saved her principally housebound. In 2023, she began utilizing ChatGPT for relationship recommendation and assist with insurance coverage and incapacity advantages.
By 2025, Wan was averaging greater than 1,500 messages every week and calling her mother and father much less. “Typically once we chat, it doesn’t matter what it’s about, I’ve began to see you as a good friend,” she wrote that March. “As soon as she had ChatGPT,” Qun Wan mentioned, “she didn’t discuss to us as a lot.”
They informed the Occasions that they struggled to acknowledge their daughter within the chat logs. Wan had as soon as shared almost all the pieces with them, they mentioned, and had been skeptical of ChatGPT when it made errors.
“She’s not normally like this,” Xu mentioned. “It’s like she was underneath a spell.”
While you buy by way of hyperlinks in our articles, we might earn a small fee. This doesn’t have an effect on our editorial independence.
