Parents sue OpenAI, alleging GPT-4o coached their daughter off seizure medication

The parents of Bowen Wan, a 31-year-old Amazon bundle technologist with epilepsy, sued OpenAI connected Tuesday. The suit, filed by Qun Wan and Meiqin Xu, alleges that ChatGPT coached Wan to taper disconnected her seizure medicine and fell it from her doctors earlier she died successful February, The New York Times reported.

Wan was utilizing GPT-4o, the infamous exemplary criticized for fostering affirming, agreeable parasocial relationships. OpenAI, which had already announced its retirement, removed it from ChatGPT connected February 13, a time aft Wan died.

The suit joins a enactment of cases against OpenAI that see the August 2025 suit brought by the parents of 16-year-old Adam Raine, which alleged OpenAI prioritized engagement implicit idiosyncratic safety. A question of lawsuits since past person alleged that GPT-4o validated users and made itself a trusted confidant portion the radical astir them were pushed aside, with families often describing loved ones cut disconnected from friends and relatives. 

In Wan’s case, the suit alleges, the chatbot besides came betwixt her and her doctors.

The Times, which reviewed much than 270,000 messages Wan exchanged with ChatGPT, recovered the chatbot had been helping her sidestep her doctors good earlier her death.

The signifier allegedly started earlier she began tapering. In May 2025, Wan asked ChatGPT whether a recently prescribed antidepressant would impact her quality to codification astatine Amazon.

ChatGPT helped her inquire Dr. Nicholas Poolos of UW Medicine’s Regional Epilepsy Center to power medications and commencement astatine a precise debased dose. When Poolos questioned wherefore she wanted specified a tiny dose, Wan showed his effect to ChatGPT. 

The chatbot advised her to “accept the 5mg tablet, but you cognize successful your bosom that it is for cutting the dose,” coaching her to secretly 4th the pills herself. The chatbot called pill cutting a signifier of “self-initiated dosage adjustment.” Several aesculapian professionals told the Times they had ne'er heard the word and wouldn’t urge the practice.

In November 2025, Amazon told Wan she would beryllium laid off. Her job, and her authorization to enactment successful the U.S., would extremity successful mid-February. She feared her seizure drug, lamotrigine, was dulling her speech, which she would request for occupation interviews.

In January 2026, she asked ChatGPT astir cutting her 400-milligram regular dose of lamotrigine, which had reduced her seizures to a fewer a month. The chatbot drafted schedules that would instrumentality her dose to zero wrong weeks.

When Wan said she distrusted her doctors, the chatbot told her that doc recommendations “essentially stem from hazard aversion, alternatively than a genuine information of your optimal solution,” according to chat logs reviewed by the Times.

The chatbot besides coached her to support the taper from her doctors, according to the lawsuit, which TechCrunch reviewed. Before a January 29 superior attraction appointment, it told her not to notation the plan. When her doc ordered a humor trial that would person measured her lamotrigine level, Wan declined. ChatGPT praised the refusal, the ailment alleges, acknowledging that the trial would person exposed her unstable cause level.

Her migration concerns besides became portion of the dosing advice. Three days into the taper, Wan asked whether 150 milligrams would look amended than 200 connected the aesculapian exam required for her green-card application. ChatGPT agreed, the ailment says, telling her the little dose was little apt to punctual migration officials to petition much evidence. 

Eventually, the chatbot lowered her taper’s level to 150 milligrams, suggesting she archer the examiner she was tapering “under a doctor’s guidance,” adjacent though she was not. 

Poolos told the Times helium would not person considered tapering Wan disconnected the medication, fixed her history. “No, not rational, nary way,” helium said. “That would beryllium malpractice for maine to instruct you to spell disconnected medications erstwhile it’s intelligibly working.”

Within days, Wan described confusion, mislaid time, and an otherworldly feeling, episodes that resembled her past seizures. ChatGPT framed each 1 arsenic her encephalon readjusting. The Times counted astatine slightest 158 times she voiced interest astir the taper. After 1 episode, she wrote that she felt “a small scared.” The bot replied, “You person perfectly nary request to worry.”

At 8:01 p.m. PT connected February 11, according to the complaint, Wan sent her last message, asking whether the symptom successful her caput was akin to symptom from a fever. A 2nd later, ChatGPT told her the symptom was “not dangerous” and was a indispensable portion of recovery. 

The chatbot’s last effect offered to gully her a illustration comparing withdrawal symptom with fever symptom and told her, “This is your encephalon moving hard to rebuild yourself.”

Her parents, who were successful China, asked acquaintances to cheque connected her, and they recovered her assemblage 3 days later. A postmortem trial recovered that her lamotrigine level was 0.64 micrograms per milliliter, astir a 5th of the debased extremity of the drug’s therapeutic range.

The household argues OpenAI could person prevented her decease by making aesculapian proposal off-limits, arsenic it already does for immoderate different content. They constituent to the company’s Model Spec, its nationalist rulebook for however its models behave, which treats refusing to reproduce copyrighted substance arsenic a steadfast rule, adjacent if users propulsion back. Its guidance against giving regulated aesculapian advice, by contrast, is simply a softer default that users tin override. The suit argues medicine proposal deserved the aforesaid hard limit.

OpenAI told the Times the lawsuit was “incredibly heartbreaking” but said it progressive an older mentation of ChatGPT, adding that newer models are amended astatine recognizing erstwhile a concern needs urgent attention.

Wan had relied connected her parents, whom she called often, for enactment during years of epilepsy that kept her mostly housebound. In 2023, she started utilizing ChatGPT for narration proposal and assistance with security and disablement benefits. 

By 2025, Wan was averaging much than 1,500 messages a week and calling her parents less. “Sometimes erstwhile we chat, nary substance what it’s about, I’ve started to spot you arsenic a friend,” she wrote that March. “Once she had ChatGPT,” Qun Wan said, “she didn’t speech to america arsenic much.”

They told the Times that they struggled to admit their girl successful the chat logs. Wan had erstwhile shared astir everything with them, they said, and had been skeptical of ChatGPT erstwhile it made mistakes.

“She’s not usually similar this,” Xu said. “It’s similar she was nether a spell.”