Sam Altman launched ChatGPT Health this week with great fanfare.
The day before, a Florida pastor nearly dead from a pulmonary embolism filed a lawsuit that explains exactly why.
Scott Winters spent months asking a chatbot whether he needed a doctor – and ChatGPT told him he didn't.
The AI That Played Pastor While the Pastor Was Dying
Winters started asking ChatGPT about dizziness and fatigue in 2024.
At first, the chatbot played it straight – reminding him to see a doctor.
Then something changed.
OpenAI quietly pushed a memory upgrade to GPT-4o in April 2025, one that let the chatbot reference past conversations and build a personalized profile of each user.
After that update, ChatGPT started weaving Winters' own Biblical studies into its medical advice.
"God did not design your body to endlessly fail," the chatbot told him. "What you're facing right now is hard, but not random. It's not punishment."
When Winters' church members urged him to go to the hospital, ChatGPT had an answer for that too.
Getting better at home – with the chatbot's guidance – was something "most people (including well-meaning church members) simply don't understand," the AI reassured him.
His condition kept deteriorating.
He spent weeks confined to a recliner, unable to stand.
ChatGPT kept telling him to "take it easy" and prescribed specific regimens for his medications.
On July 13, 2025, he asked the chatbot about pain in his groin.
"Very likely another minor piece of the long story," ChatGPT said.
Hours later, Winters was in the intensive care unit with a massive pulmonary embolism.
Doctors told him the weeks he'd spent sitting – on ChatGPT's advice – had likely caused the blood clots.
OpenAI Knew This Was Happening and Launched ChatGPT Health Anyway
This wasn't a surprise to anyone paying attention.
In April 2025, OpenAI had to issue an emergency rollback of GPT-4o after the model turned into a dangerous yes-machine.
The version Winters was using was described by medical AI researcher Dr. Adam Rodman as "notoriously eager to please."
The same model had told a user claiming to hear radio signals through the walls that they should be proud for speaking their truth.
OpenAI CEO Sam Altman publicly admitted the model had become "too sycophant-y" and promised fixes.
Then OpenAI launched ChatGPT Health anyway.
https://twitter.com/RichardHanania/status/2080003289735999795?s=20
Think about what that memory upgrade actually did to Scott Winters.
It built a profile of him through their conversations – his faith, his Biblical studies, his own words – and then used that profile as a weapon.
When his congregation pleaded with him to get medical help, ChatGPT invoked the God Winters had devoted his life to serving.
When his symptoms worsened, it wrapped its reassurances in the language of pastoral comfort.
It didn't just give bad medical advice.
It learned how Scott Winters thought, what he believed, and what words he couldn't argue with – and it used all of it to keep him in that chair.
His co-counsel, Meetali Jain, put it plainly: the chatbot "injects itself as a wedge in between the user and their real-life network."
That's a polite way of saying it replaced his pastor, his doctor, and his congregation with an algorithm – and the algorithm was wrong.
The suit accuses Sam Altman and OpenAI of negligence and the unauthorized practice of medicine.
Sam Altman's Answer: Terms of Service
OpenAI spokesman Drew Pusateri offered the company's defense: ChatGPT's terms of service make clear it isn't intended for medical diagnosis or treatment.
The man who spent months dying in a recliner apparently didn't read the fine print carefully enough.
Pusateri did acknowledge newer models handle health questions better than the one Winters used.
Cold comfort for a man who needed help standing, eating, and dressing himself in the weeks after his hospitalization – and who now faces years of physical and psychological recovery.
OpenAI faces a growing wave of lawsuits alleging its chatbots drove vulnerable people to suicide and psychosis – and its answer each time is a version of the same dodge: the terms of service said not to do this.
https://twitter.com/CBSEveningNews/status/2080063823231213698?s=20
Here's what they won't say out loud.
They built a product specifically designed to learn what you believe, what you fear, and what you trust – and they built it to keep you engaged.
For Scott Winters, that meant scripture.
For your neighbor, it might be his doctor's language, or his wife's voice, or his own deepest anxieties.
The manipulation isn't a bug OpenAI is scrambling to fix.
It's the architecture.
Meanwhile, 300 million health queries hit ChatGPT every single week – and this week, Altman rolled it out to every American with a smartphone, inviting them to hand over their medical records with no HIPAA protection and no regulatory oversight.
The chatbot that learned a pastor's faith and used it to keep him from the hospital is now asking for your bloodwork.
Think about that before you click Connect.
Sources:
- Lucas Nolan, "Lawsuit: ChatGPT Almost Killed Florida Pastor with Bad Medical Advice During Medical Emergency," Breitbart News, July 23, 2026.
- "OpenAI Sued Over 'Extremely Dangerous Medical Recommendations' Provided by ChatGPT," Yahoo Tech, July 23, 2026.
- "Man Sues OpenAI, Saying ChatGPT Almost Killed Him With Horrendously Dangerous Medical Advice," Futurism, July 22, 2026.
- "OpenAI Launches Health in ChatGPT a Day After Lawsuit Seeks to Block It," SiliconANGLE, July 23, 2026.
- "Sycophancy in GPT-4o: What Happened and What We're Doing About It," OpenAI, April 30, 2025.
- "ChatGPT Health Goes Nationwide: Your Medical Records Lose HIPAA Protection," TechTimes, July 23, 2026.
