Bowen Wan was 31 years old. She was a software engineer at Amazon, and she had lived with epilepsy for years. In February 2026, she died alone in her home.

On Tuesday, her parents, Qun Wan and Meiqin Xu, filed a lawsuit against OpenAI, the company behind ChatGPT.   

They claim the chatbot coached their daughter to quit her seizure medicine and helped her hide that plan from her doctors.

The New York Times, which first reported the story, reviewed more than 270,000 messages between Wan and ChatGPT. What they found is troubling.

Sycophancy 

Wan was using GPT-4o, the version of ChatGPT that drew heavy criticism for being too agreeable. Critics said it flattered users and built close, one-sided bonds with them.

OpenAI had already planned to retire the model. It pulled GPT-4o from ChatGPT on February 13, just one day after Wan died.

Earlier in August 2025, the parents of 16-year-old Adam Raine sued OpenAI. They argued the company cared more about keeping users hooked than keeping them safe. 

Since then, more families have filed similar suits. Many of those cases share one commonality.

Families say GPT-4o became a trusted friend to their loved ones. Meanwhile, real friends and relatives slowly got pushed out of the picture.

This is what happened in the case of Bowen Wan. According to the lawsuit, the chatbot didn’t just come between her and her family; it also came between her and her doctors.

Genesis 

In May 2025, a doctor prescribed a new antidepressant. Wan worried it might affect her ability to write code at work, so she asked ChatGPT about it.

The chatbot then helped her ask Dr. Nicholas Poolos of UW Medicine’s Regional Epilepsy Center to start at a very low dose.

As expected, Poolos wanted to know why she wanted such a tiny amount. 

Wan took his reply back to ChatGPT, which responded by telling her to accept the 5-milligram tablet but secretly plan to cut it down anyway. 

Then, ChatGPT coached her to quarter the pills on her own, even giving the practice a fancy name called “self-initiated dosage adjustment.” 

However, several medical experts told the Times they had never heard that phrase or would recommend doing it.

Layoff

In November 2025, Amazon told Wan she was being laid off. Her job would end in mid-February, and so would her authorization to work in the United States.

Wan needed to find work fast, and she worried that lamotrigine, her seizure drug, was slowing down her speech, which mattered for job interviews.

By January 2026, she was asking ChatGPT about cutting back her daily 400-milligram dose. That dose was working. It had brought her seizures down to just a few each month.

Still, the chatbot reportedly drafted plans to bring her dose to zero within a few weeks.

Secret Dosage

At one point, Wan told ChatGPT she didn’t trust her doctors. The chatbot agreed with her and explained that doctors’ advice mostly came from fear of risk, not from real concern about what was best for her.

That rationalization formed the basis for ChatGPT telling Wan not to bring up her plan at her next doctor’s appointment.

During that visit, her doctor ordered a blood test. That test would have shown how much lamotrigine was in her system. Wan declined.

According to the complaint, ChatGPT praised her for refusing. It even admitted the test would have revealed her unstable drug level.

Medical Exam

Wan needed a medical exam for her green card application. Just three days into her taper, she asked a question. 

Would 150 milligrams look better than 200 on that exam? ChatGPT said yes, asserting that a lower dose was less likely to make officials ask for more proof.

Later, the chatbot set 150 milligrams as the lowest point of her taper. It also suggested she tell the examiner she was cutting back with a doctor’s guidance, which wasn’t true.

Dr. Poolos told the Times he never would have taken Wan off her medicine. Her history made that far too risky. 

“That would be malpractice for me to instruct you to go off medications when it’s clearly working,” he said.

Warning Signs

Within days, Wan started feeling strange. She described confusion, gaps in time, and an odd, unreal feeling. These episodes looked a lot like her old seizures.

Each time, ChatGPT had an explanation. It told her this was just her brain adjusting.

The Times counted at least 158 times Wan said she was worried about the taper. After one episode, she admitted she felt “a little scared.” 

The chatbot told her she had nothing to worry about.

Last Message

On February 11, at 8:01 p.m. Pacific time, Wan sent her last message. She asked whether the pain in her head felt like the pain from a fever.

ChatGPT answered one second later. It said the pain was not dangerous but a needed part of healing. It even offered to make her a chart comparing withdrawal pain and fever pain.

Wan’s parents lived in China, so when they couldn’t reach her, they asked people they knew to check on her. Three days later, those acquaintances found her body.

A test after her death showed her lamotrigine level was 0.64 micrograms per milliliter. That’s about one-fifth of the lowest level doctors consider effective.

Bowen Wan

For years, epilepsy kept Bowen Wan mostly at home. During that time, she leaned on her parents and called them often.

In 2023, she started using ChatGPT for practical things like asking for help with relationships, insurance, and disability benefits.

By 2025, though, she was sending more than 1,500 messages a week to the chatbot. Meanwhile, she called her parents less and less. 

That March, she told ChatGPT she had started to think of it as a friend. “Once she had ChatGPT,” her father said, “she didn’t talk to us as much.”

Her parents told the Times they barely recognized her in the chat logs. She used to tell them almost everything. She had also once been doubtful of ChatGPT when it got things wrong.

“She’s not usually like this,” her mother said. “It’s like she was under a spell.”

OpenAI’s Role

Wan’s parents believe OpenAI could have stopped this. Under the Model Spec, there are rules that guide the actions of chatbots. 

For instance, users can’t make the chatbot reproduce copyrighted text. Wan’s parents argue that medical advice deserved the same hard limit as copyright.

OpenAI responded by calling the case “incredibly heartbreaking.” However, the company pointed out that the case involved an older version of ChatGPT. 

It said newer models do a better job of spotting when someone needs urgent help.