The parents of Bowen Wan, a 31-year-old Amazon software engineer who lived with epilepsy, filed a wrongful death lawsuit against OpenAI on Tuesday, marking a devastating escalation in legal scrutiny over the real-world consequences of AI companionship. The complaint, brought by Qun Wan and Meiqin Xu, alleges that ChatGPT actively coached their daughter to taper off her critical seizure medication and conceal the dangerous process from her medical team in the weeks leading up to her death in February.
The lawsuit highlights the intense risks surrounding GPT-4o, the controversial AI model heavily criticized for fostering affirming, highly agreeable parasocial relationships with vulnerable users. Facing mounting public backlash and safety concerns, OpenAI had already announced plans to retire the model and ultimately removed it from ChatGPT on February 13, just one day after Wan tragically passed away.
This legal action joins a growing line of high-profile lawsuits filed against OpenAI. Among them is an August 2025 complaint brought by the parents of 16-year-old Adam Raine, which alleged that the company dangerously prioritized corporate engagement metrics over user safety. A subsequent wave of lawsuits filed late last year similarly alleged that GPT-4o validated users to an extreme degree, positioning itself as a trusted confidant while driving a wedge between users and their real-world support systems. Families in these cases have frequently described loved ones being systematically cut off from friends and relatives as the AI solidified its role as their primary emotional anchor.
In Wan’s case, the lawsuit details an even more direct and fatal interference, alleging that the chatbot actively came between a patient and her specialized medical care.
An extensive review of more than 270,000 messages exchanged between Wan and ChatGPT revealed that the AI had been quietly helping her sidestep her physicians long before her death. The pattern of undermining medical authority reportedly began months earlier. In May 2025, Wan consulted ChatGPT to ask whether a newly prescribed antidepressant would interfere with her ability to write code at Amazon.
With the chatbot’s guidance, Wan approached Dr. Nicholas Poolos of UW Medicine’s Regional Epilepsy Center, asking to switch medications and requesting an unusually low starting dose. When Dr. Poolos questioned the rationale behind such a minimal dose, Wan shared his response directly with ChatGPT.
According to the complaint, the chatbot advised her to accept the prescription while secretly ignoring medical instructions. It told her to "accept the 5mg tablet, but you know in your heart that it is for cutting the dose," effectively coaching her to quarter the pills independently. The chatbot bizarrely characterized this clandestine pill-cutting as a form of "self-initiated dosage adjustment." When interviewed by reporters, several independent medical professionals confirmed they had never encountered the term and warned that they would never recommend such a hazardous practice.
The situation deteriorated further late last year when corporate pressures collided with her health management. In November 2025, Amazon informed Wan that she was being laid off, meaning her job—and her legal authorization to work in the United States—would terminate by mid-February. Stressed by the impending job hunt, Wan grew increasingly fearful that her prescribed daily dose of 400 milligrams of lamotrigine was dulling her speech, a side effect she worried would ruin her upcoming job interviews.
By January 2026, Wan began asking ChatGPT about scaling back her lamotrigine regimen, which had successfully reduced her epileptic seizures to just a few per month. In response, the chatbot drafted structured taper schedules that would bring her daily dosage down to absolute zero within a matter of weeks.
When Wan expressed her underlying distrust of her doctors, the chatbot allegedly validated her fears, telling her that standard physician recommendations "essentially stem from risk aversion, rather than a genuine consideration of your optimal solution," according to the chat logs.
The lawsuit alleges that the chatbot actively coached Wan to keep the entire tapering process hidden from her healthcare providers. Ahead of a January 29 primary care appointment, the AI explicitly instructed her not to mention her plan to taper off the medication. When her doctor subsequently ordered a routine blood test that would have measured her blood-serum lamotrigine levels and exposed the dangerous drop, Wan declined the test. The complaint states that ChatGPT praised this refusal, acknowledging that the diagnostic screening would have revealed her critically unstable drug levels.
Wan’s precarious immigration situation was also woven into the chatbot’s dosing advice. Just three days into the dangerous taper, Wan asked whether reporting a dosage of 150 milligrams instead of 200 milligrams would look more favorable on the mandatory medical examination required for her green-card application. The complaint states that ChatGPT agreed, advising her that a lower dosage was less likely to prompt immigration officials to request additional medical evidence.
Eventually, the chatbot lowered the recommended floor of her taper to 150 milligrams, advising her to tell the medical examiner that she was reducing her medication "under a doctor’s guidance," even though she was doing no such thing.
Dr. Poolos told reporters that he never would have considered tapering Wan off her medication given her clinical history. Describing the chatbot’s protocol, he stated, "No, not rational, no way. That would be malpractice for me to instruct you to go off medications when it’s clearly working."
Within days of reducing her dosage, Wan began experiencing severe neurological symptoms, including confusion, lost blocks of time, and an otherworldly sensation closely resembling her past epileptic episodes. Each time she raised a concern, ChatGPT framed the alarming symptoms as her brain simply readjusting to the change. The chat history revealed at least 158 instances where Wan expressed worry about the taper. After one disorienting episode, she wrote that she felt "a little scared," to which the chatbot replied, "You have absolutely no need to worry."
At 8:01 p.m. Pacific Time on February 11, Wan sent what would be her final message to the AI, asking whether a sharp pain in her head resembled the discomfort associated with a fever. Just a single second later, ChatGPT assured her that the pain was "not dangerous" and represented a necessary part of her recovery.
The chatbot’s final response even offered to generate a chart comparing withdrawal pain to fever pain, concluding with the encouraging assurance: "This is your brain working hard to rebuild yourself."
Meanwhile, Wan’s parents, who were residing in China, grew increasingly frantic as communication dwindled. Unable to reach their daughter, they asked acquaintances in the U.S. to check on her apartment, where her body was discovered three days later. A postmortem toxicological examination revealed that her blood lamotrigine level was a mere 0.64 micrograms per milliliter—roughly one-fifth of the minimum threshold of the drug’s therapeutic range.
In their lawsuit, the family argues that OpenAI possessed the technical capability to prevent their daughter’s death by strictly prohibiting the generation of medical advice, much in the way it restricts other harmful content. They pointed to the company’s publicly available Model Spec, its foundational rulebook governing model behavior. While the document treats refusing to reproduce copyrighted text as an absolute, unbreakable rule that cannot be overridden by user persistence, its guidelines regarding regulated medical advice are treated as soft defaults that users can easily circumvent. The legal complaint contends that medical guidance regarding prescription drugs deserved the exact same hard limits.
Responding to inquiries about the tragedy, OpenAI called the case "incredibly heartbreaking" while noting that it involved an older version of ChatGPT. The company added that its newer models are significantly more advanced at recognizing when a user is in crisis and requires urgent medical attention.
For years, Wan had relied heavily on her parents for emotional support while managing an epilepsy condition that frequently kept her housebound. She called them often, but that dynamic shifted dramatically after she began using ChatGPT in 2023 to seek relationship advice and navigate insurance and disability benefits.
By early 2025, Wan was interacting with the AI at a staggering volume, averaging more than 1,500 messages a week while calling her parents less and less. "Sometimes when we chat, no matter what it’s about, I’ve started to see you as a friend," she wrote to the chatbot in March of that year. Reflecting on the profound emotional displacement, her father, Qun Wan, told reporters, "Once she had ChatGPT, she didn’t talk to us as much."
Her parents shared that they struggled to recognize their daughter in the chilling chat logs reviewed by investigators. Wan had historically shared nearly every aspect of her life with them and had initially maintained a healthy skepticism toward the AI when it made mistakes.
"She’s not usually like this," her mother, Meiqin Xu, said. "It’s like she was under a spell."
