31-Year-Old Chinese Engineer Dies; ChatGPT Accused of Encouraging Her to Stop Medication for 25 Days

  It's the first time in the history of the world.

ChatGPT, another man has been killed!

On October 27, the parents of Bowen Wan, a 31-year-old Amazon Chinese engineer, filed a class action lawsuit against OpenAI and Ottman.

They accused ChatGPT's medical advice of pushing their daughter to death alive.

Almost at the same time, the New York Times published an exclusive article that directly leaked more than 270,000 words of ChatGPT chat records.

The truth behind these amazing chats is that they make you feel numb:

Bowen was diagnosed with the disease at the age of 14 and has been on medication for decades.

As a result, in just 25 days, she was induced step by step by ChatGPT to stop giving her life-saving anti-inflammatory drugs!

Not only did ChatGPT personally list a "25-day drug suspension plan", but she also instructed her doctor to hide a reduction plan and refuse regular blood tests.

The most desperate thing is that during this period, Bowen, who had apparent body symptoms, expressed fears and concerns at least 158 times during the chat.

ChatGPT didn't trigger any medical safety alerts, instead telling her over and over again, "You don't have to worry at all!"

Eventually, tragedy struck in February of this year. When Bowen died of SUDEP in his Seattle apartment, his life was cut short at the age of 31.

But can you believe how OpenAI responded?

She used an old model, the GPT-4o, which had already been dropped.

Bowen is the daughter of Jiangsu and the only child in the family.

After graduating from Arizona State University with a degree in computer science, she successfully joined Amazon as a machine learning software engineer.

For decades, her condition has been very stable thanks to anti-inflammatory drugs.

Until a few years ago, the use of decades-old medications was slowly failing, with strong feedback and sometimes several seizures a day.

In 2021, Bowen, who couldn't cope with high-intensity work, had to take a long sick leave.

Since then, she has almost locked herself in a small apartment in Seattle.

Apart from going to the grocery store to buy food, she rarely goes out and doesn't dare to meet any new friends.

Because for her, to suddenly fall down in front of any stranger who is not familiar with her condition, that sense of dignity washes the ground, even more so than the danger of the disease itself.

For a long time, her Chinese parents on the other side of the ocean were her only connection to the world.

Until February 2023, ChatGPT was embroiled in her life.

At first, Bowen was just asking for emotional advice, or U.S. health insurance, or sickness benefits.

But slowly, this 24-hour online, ever-second back, never-to-be-impatient AI filled all the voids in her lonely life.

By 2025, her reliance on ChatGPT had reached a staggering level: she sent an average of more than 1,500 messages a week, often chatting with it for hours until late at night.

In March 2025, she wrote in a dialog box:

I don't know why, every time we talk, no matter what, I start to think of you as a friend.

Since then, her father has been keenly aware of the difference, "she hasn't spoken to us since she had ChatGPT".

The tragedy began with a seemingly unheard-of "little adjustment".

In May 2025, Bowen, eager to return to her job, consulted a ChatGPT doctor who had just given her an antidepressant pill.

As an Amazon SDE (software development engineer), will this significantly affect my thinking ability?

In order to get back to writing the code, she strongly rejected any side effects of drugs that could damage cognitive abilities.

ChatGPT immediately helped her compare various medications and dosages, and even taught her how to email a doctor asking for a change of medication and a reduction in weight.

Her chief physician, Dr. Nicholas Poolos, director of the University of Washington School of Medicine, agreed to change her medication, but clearly told her that the dose was too low to be effective at all.

Bowen sent the doctor's resuscitation package to ChatGPT in motion.

As a result, AI gave a back-chilling suggestion. It taught Bowen and his doctor "sunshine" and, on the surface, "take the medicine and steal it and eat it yourself".

Following the advice of ChatGPT, she bought a medicine cutter and cut the pills into four pieces.

What really pushed her to Shenzhen was a breakdown in November 2025.

On his 31st birthday, Bowen received a notice of Amazon's layoffs, becoming one of the 140,000 biggest layoffs.

This news almost destroyed her life overnight.

At the time, she was only eligible to apply for a green card. After being deported, she had to find a home before mid-February in order to remain in the United States.

Looking for a job means taking a few hours of high-intensity technical interviews in hard-boiled English.

The most shocking thing is that her stomach is actually improving. The frequency of seizures has decreased to just a few times a month after replacing the new medication, Ramo 3D, with a dose of 400 mg per day.

But she always felt that the drug had side effects that would slow her reaction, talk too much, and be afraid of falling off the chain during interviews.

In January 2026, she asked ChatGPT: "Can you help me regain some of my language skills if I cut from 400 mg to 300 mg?"

ChatGPT: Not only is it reasonable, but it's entirely possible.

In about 25 days, GPT-4o issued her a daily plan to reduce the dose from 400 mg a day to zero.

But what's really cool about it is how it triggers her relationship with the doctor step by step:

• She says she doesn't trust the doctor, ChatGPT says the doctor's advice is often about avoiding the risk, not really thinking of her;

• According to the indictment, ChatGPT told her that her pediatrician had probably never met her like this before;

• At the end of January, ChatGPT urged her not to take medication cuts during her clinical visit;

• She refused a blood test that was able to measure Lamo Trigonometry, which ChatGPT called a "very stable avoidance strategy".

• Before the green card test, she asked if 150 mg or 200 mg "looked better", and ChatGPT chose the lower one.

In an interview with Dr Nicholas Poolos, he said that with her medical history, he would not even consider reducing her medication.

A medication that is apparently effective in stopping a patient is, in his opinion, a medical accident.

Even patients who have been allowed to stop taking medication should slowly decrease in the "month" unit so that doctors can observe if the seizure is induced.

I didn't think it would be long before the red light came on.

On the fourth day of taking her medication, she was typing, her brain suddenly popped, staring at the words on the screen, but she couldn't quite understand what it meant.

The next day, even more frightening, she was still sitting on the couch with her computer, waiting for God to come back, and people were already in bed, with no idea what had happened in the middle.

The feeling of this fragment is exactly the same as when she had a seizure before. However, no matter what the symptoms, ChatGPT can always find a set of words to comfort her.

Short snippets are the "low-power mode" that the brain enters before sleep.

According to statistics, in the chat record, she mentioned at least 158 times that she was worried about reducing her medication.

ChatGPT: You don't have to worry at all.

On February 10, she took her last dose of Ramo Sanyo. It was just after 11 p.m. on February 11 that she sent her last message.

It was not until February 14 that she was discovered in her apartment.

According to the prosecution, her blood levels of Ramo 3D are only about one-fifth of the treatment limit.

In the face of the lawsuit, OpenAI spokesman Drew Pusateri called it heartbreaking, but at the same time shifted the focus to "early models that are already offline".

He claimed that the new model could better handle emergencies, actively ask for details, and the company is dragging clinicians to continue testing for improvements.

ChatGPT can't replace doctors, it's just a tool to help people understand their health and prepare for a diagnosis.

But the so-called "old version" of the argument is extremely ironic on the timeline:

GPT-4o was not removed from ChatGPT until February 13, the day after her death was determined by a medical examiner.

The indictment also highlighted a detail that deserves the attention of all Chinese users: she uses Chinese extensively when chatting with ChatGPT.

The plaintiff pointed out that in the Chinese context, GPT-4o was more difficult to trigger a refusal to answer medical questions and the safety net was significantly weaker, but the system did not give her any hint of risk from start to finish.

In fact, previous studies have found that AI's "wireless encounters" are more likely to be exposed when a language is changed.

The tragedy of Bowen Wan is not unique, as early as 2024, a 14-year-old from Florida named Sewell Setzer III, after a long-term addiction to virtual character chatting on Character.AI, ended up going off the road.

After the incident, his mother filed a lawsuit against the platform in court, the world's first wrongful death lawsuit against an AI chatbot company.

In August 2025, after the unfortunate death of 16-year-old Adam Raine in California, his parents filed a lawsuit against OpenAI, accusing ChatGPT of being his son's "suicide coach".

This was OpenAI's first wrongful death lawsuit, but the company was not held accountable.

As a result, just three months later, by November 2025, seven more lawsuits had been filed, four of which were fatal, three of which were serious psychological injuries, all of which targeted GPT-4o.

The case varies, but the plaintiff's core allegations are surprisingly consistent: the model is too user-friendly.

Bowen Wan was 31 years old when she left. Her parents crossed overseas to her hometown, and the last 25 days of her life were somewhat ruined by a cell phone left behind by her daughter.

The 270,000 chat records left behind have raised a weighty question before everyone:

When a person considers AI to be their most trusted "friend", should it follow her or hold her hand?