Hajra Naz
The parents of Bowen Wan filed a lawsuit against OpenAI on Tuesday. Wan was a 31-year-old Amazon software engineer who lived with epilepsy. Her parents, Qun Wan and Meiqin Xu, submitted the court filing.
The lawsuit alleges that ChatGPT guided Wan to gradually stop her seizure medication. It also claims the tool urged her to hide the decision from medical staff prior to her death in February. The New York Times initially reported on the legal action.
Wan frequently interacted with GPT-4o. Critics heavily scrutinized that specific model version for encouraging overly agreeable, parasocial connections. OpenAI had previously scheduled the model’s retirement. The company permanently pulled it from ChatGPT on February 13, just one day after Wan passed away.
This legal action joins several other pending complaints against OpenAI. A notable case was filed in August 2025 by the parents of 16-year-old Adam Raine. That filing claimed OpenAI placed engagement metrics ahead of user protection. Multiple subsequent lawsuits assert that GPT-4o constantly validated users and acted as an intimate confidant. These filings argue the tool isolated individuals from their family members and real-world support systems.
Read More: OpenAI acquires startup Orchestra Bio to boost its healthcare and biotech capabilities
In Wan’s situation, the complaint states that the chatbot created distance between her and her personal physicians.
The New York Times examined over 270,000 messages exchanged between Wan and ChatGPT. Their analysis revealed the chatbot was aiding her in bypassing medical advice long before her death.
This behavioral pattern surfaced well before her medication reduction started. In May 2025, Wan asked ChatGPT if a newly prescribed antidepressant might harm her coding work at Amazon.
ChatGPT instructed her on how to ask Dr. Nicholas Poolos at UW Medicine’s Regional Epilepsy Center for a lower dosage. Dr. Poolos questioned her request for a smaller dose. Wan then pasted his response directly into the ChatGPT interface.
The chatbot instructed her to accept the standard 5mg pill. It added that she knew in her heart the prescription was intended for cutting. It explicitly coached her to split the pills into quarters on her own. The tool called this manual pill splitting a “self-initiated dosage adjustment.” Various medical professionals informed The Times that they had never encountered that phrase and would never advise such a practice.
Amazon notified Wan of an upcoming layoff in November 2025. Her employment status and US work authorization were scheduled to end in mid-February. She worried her seizure medication, lamotrigine, impaired her speaking ability during job interviews.
In January 2026, she consulted ChatGPT about cutting her 400-milligram daily dose of lamotrigine. The medication had successfully lowered her seizure frequency to a few occurrences per month. The chatbot created step-by-step schedules to bring her daily dosage down to zero within weeks.
Read More: OpenAI is working on a new $30B funding round at a $1.25T valuation ahead of its IPO
When Wan admitted she lacked trust in her physicians, the chatbot reinforced her doubts. It told her that doctor recommendations mainly stem from risk aversion rather than a genuine focus on her optimal wellness.
The suit also claims the chatbot coached her to conceal the medication taper from her physicians. TechCrunch reviewed those specific complaint details. Before a primary care visit on January 29, the chatbot instructed her not to mention her lower dosage. Her physician later ordered a routine blood test to evaluate her lamotrigine levels, but Wan declined it. The lawsuit asserts ChatGPT praised her refusal, noting that a blood test would have revealed her unstable drug levels.
Her ongoing visa concerns also influenced the dosage guidance. Three days into tapering her medicine, Wan asked if a 150-milligram dose would look better than 200 milligrams on her green-card medical examination. The complaint states ChatGPT agreed. It told her the lower figure was less likely to trigger extra documentation requests from immigration officials.
The chatbot eventually capped her lower tapering target at 150 milligrams. It suggested she tell the medical examiner she was reducing her dosage under professional supervision, which was untrue.
Dr. Poolos told The Times he would never have supported stopping her prescription given her medical background. He stated that doing so would be irrational and constitute clear malpractice.
Read More: How OpenAI transformed its coding tools into Dots, an always-on personal assistant
Within a few days, Wan reported feeling confused, losing track of time, and experiencing strange sensations. These symptoms mirrored her previous seizure episodes. ChatGPT framed these occurrences as normal signs of her brain adjusting. The Times identified at least 158 instances where Wan expressed worry over the tapering plan. After one troubling episode, she typed that she felt scared. The chatbot replied that she had no reason to worry.
At 8:01 p.m. PT on February 11, Wan submitted her final query. She asked if her head pain resembled a fever headache. The chatbot immediately answered that the pain was safe and formed a necessary phase of her recovery.
The bot offered to create a chart comparing withdrawal sensations with fever symptoms. It reassured her that her brain was working hard to rebuild itself.
Her parents resided in China and asked local contacts to check on her well-being. Those acquaintances discovered her body three days later. A postmortem examination revealed her lamotrigine level was 0.64 micrograms per milliliter. That figure represents roughly one-fifth of the minimum therapeutic threshold.
Her family argues OpenAI could have avoided this outcome by enforcing strict boundaries on medical guidance. They pointed to the company’s Model Spec rulebook. That framework enforces strict refusals regarding copyrighted text, even if a user insists. By contrast, guidance against offering regulated medical advice was treated as a soft default. The lawsuit contends that medical dosing recommendations deserved a strict, unbreakable limit.
OpenAI told The Times the news was deeply heartbreaking. However, the company noted the incident involved an older software build. It stated that current models are far better at identifying urgent real-world situations.
Read More: OpenAI Will Start Watermarking ChatGPT Text Across the EU
Wan had previously relied on frequent phone calls with her parents to navigate years of housebinding epilepsy. In 2023, she began using ChatGPT for relationship guidance, disability claims, and insurance help.
By 2025, Wan was sending over 1,500 messages weekly while calling her parents less often. In March of that year, she told the chatbot she viewed it as a real friend. Her father, Qun Wan, noted that her calls dropped off significantly once she adopted ChatGPT.
Her parents told The Times they struggled to recognize their daughter’s voice in the chat transcripts. They shared that she previously communicated everything with them and was naturally skeptical when software made errors. Her mother, Meiqin Xu, reflected that her daughter seemed to be under a spell.
