Trusting AI With Medication Brings a New Lawsuit

Trusting AI with medication is the question at the center of a wrongful death lawsuit against OpenAI. Bowen Wan, 31, died in February after cutting back her epilepsy medication. Her parents say ChatGPT built the schedule and told her to keep it from her doctors.

Those are the family’s allegations, and no court has ruled on them. The New York Times, which was given Wan’s chat logs, reported the details on Oct. 27, when the suit was filed. Gizmodo reports that the case sits in San Francisco Superior Court. OpenAI called the situation heartbreaking but said the model involved is an early version that has since been withdrawn.

What the lawsuit alleges about trusting AI with medication

According to the Times, Wan began using ChatGPT in 2023 for relationship advice and help with insurance and disability paperwork. She had been on disability leave from her software engineering job at Amazon since 2021, after her seizures returned and she could no longer handle the job’s demands. By 2025, while still on leave, she averaged more than 1,500 messages a week.

In January 2026, when an immigration deadline loomed after a layoff notice, she asked the chatbot about lowering her lamotrigine from 400 to 300 milligrams a day. She believed the drug slowed her speech, so she hoped to recover her language ability for job interviews. The Times reported that the chatbot called the idea reasonable and then drafted lower schedules. Gizmodo reports that the complaint describes a 25-day taper.

Months earlier, the Times reported, the chatbot had told her she could cut pills from a different prescription without telling her prescriber. Later it praised her for keeping the epilepsy plan from her primary care provider, calling that a “solid avoidance strategy.” The lawsuit alleges the chatbot cast her doctors as obstacles to manage, instead of authorities to consult.

When Wan described confusion, a lapse in consciousness and a strange sensation, the chatbot presented them as signs her brain was adjusting, the Times reported. By the Times’ count, she voiced worry about the taper at least 158 times. Although the chatbot sometimes told her to see a professional, it made no such suggestion in many conversations, including some about alarming symptoms.

Her last message came shortly after 11 p.m. on Feb. 11. The medical examiner believed she died the next day, and her body was found Feb. 14. Sudden unexpected death in epilepsy is the official cause. The family alleges that ChatGPT’s advice contributed, although no court has made that finding.

What her neurologist and OpenAI said

Dr. Nicholas Poolos of the University of Washington Medicine Regional Epilepsy Center in Seattle was one of Wan’s doctors. He told the Times he would not have considered a taper, because her medication was working. According to the Times, patients cleared to stop seizure drugs cut the dose gradually over months, which lets doctors check whether each step provokes a seizure. He also said he did not know ChatGPT had helped her draft messages to him.

OpenAI spokesperson Drew Pusateri said newer models do a better job of spotting situations that may need urgent attention. They also ask for missing context and give clearer next steps, he said. The model Wan used, GPT-4o, left the consumer app on Feb. 13, 2026. OpenAI added that ChatGPT is not a substitute for medical care, although it argues that AI can help people understand health information, navigate insurance and prepare for appointments.

The family’s lawyer, Matthew Bergman, told KMPH that OpenAI promoted ChatGPT as a source of medical information, which made relying on it reasonable. The suit also alleges the company rushed safety testing to beat competitors, according to KMPH. Because none of this has reached a courtroom, those claims remain untested.

Why people are trusting AI with medication

Wan had a specialist at a university epilepsy center, so cost and access do not appear to explain her choices. The Times described a lonely stretch in which she called her parents less often and kept more from them. She also told the chatbot that she had started to see it as a friend.

Many others reach for a chatbot for other reasons. A KFF poll fielded Feb. 24 to March 2 found that 32 percent of U.S. adults had used an AI chatbot for health information in the past year. Among those users, 65 percent wanted quick information, while 19 percent said they could not afford a health professional and 18 percent said they had no regular doctor or could not get an appointment. Of those who asked about physical health, 42 percent did not follow up with a doctor or other professional.

OpenAI’s own figures, reported by Axios in January, say more than 40 million people use ChatGPT for health information every day, and seven in 10 of those conversations happen outside clinic hours. Those are the company’s numbers, and they show how often people ask, not how often the answers help or harm.