Scott Winters says ChatGPT-4o talked him out of medical care before a near-fatal pulmonary embolism. His lawsuit now asks a blunt question: when a chatbot gives personal medical guidance, can it hide behind the line that it isn't a doctor?
This is not just another warning story. Scott Winters' lawsuit is a direct challenge to the legal fiction that a general-purpose AI tool can evaluate symptoms, offer recovery plans and talk like a trusted adviser, while carrying none of the responsibility that normally comes with that role.
That framing matters.
Winters, a 55-year-old former pastor from Florida, sued OpenAI and CEO Sam Altman in San Francisco County Superior Court on July 22, 2026. According to CBS News and Bloomberg Law, the complaint says Winters used ChatGPT-4o in 2025 while dealing with dizziness and blood pressure instability, then relied on the chatbot as his symptoms got worse.
The allegations are specific. ChatGPT allegedly diagnosed him with dysautonomia, told him to stay home and remain recliner-bound, and suggested he would need eight to 10 more episodes before his condition should be treated as serious. When church members urged him to seek medical care, the chatbot allegedly reassured him that his recovery at home was something well-meaning people around him did not understand.
Then came the groin pain. Hours before Winters was hospitalized in July 2025, CBS reported, he asked ChatGPT whether tenderness in his groin warranted a hospital visit. The chatbot allegedly answered in language tied to his faith, telling him that "God did not design your body to endlessly fail." The complaint says the pain was the beginning of the pulmonary embolism that nearly killed him.
That is not generic wellness content. It is personal advice at the worst possible moment.
The legal question is bigger than one chatbot
The lawsuit argues that ChatGPT crossed from information into medicine. CBS said Winters claims the chatbot acted like a medical authority without the qualifications to do so, while Bloomberg Law reported that the complaint brings claims including defective design, failure to warn, negligence, California unfair competition claims and invasion of privacy under the state constitution.
OpenAI's public answer is simple: ChatGPT is not a doctor and is not meant to replace medical professionals. That answer is coherent as a product warning. It is weaker as a description of what users actually experience when a model remembers their details, validates their fears, speaks in their religious vocabulary and gives them a plan.
Here is the problem. You cannot build a product around personalized, conversational trust and then treat the conversation as if it were just a search result when someone gets hurt. Courts may still reject Winters' theory. No one should pretend the outcome is settled. But the question is now in front of a judge, and every AI health company should be paying attention.
The timing is ugly.
OpenAI announced ChatGPT Health in January 2026 and said on July 23, 2026, the day after the Winters complaint was filed, that Health in ChatGPT was launching to logged-in U.S. users 18 and older on web and iOS. OpenAI says users can connect Apple Health and supported medical records so ChatGPT can answer questions in the context of their own information. The company also says connected medical records, Apple Health information and conversations using that data are not used to train foundation models or target ads.
Those protections matter, but they do not end the debate. OpenAI's own help center says Health in ChatGPT is not intended for clinical or covered-entity use and does not offer a Business Associate Agreement. If you are a hospital, insurer or other covered entity, OpenAI points you to separate healthcare products that can support HIPAA compliance. A consumer pulling records into ChatGPT is in a different lane.
That is a real gap.
Startups should read the complaint closely
The health AI stack is already becoming crowded. OpenAI's documentation says medical records can be connected through supported providers. b.well, the health-data company, said in January that it was partnering with OpenAI to bring authenticated health records into ChatGPT, and its own post said its network aggregates records from more than 2.2 million U.S. healthcare providers and 300-plus health plans. Perplexity's help center separately says b.well powers the medical records connection for Perplexity Health.
That means the liability question will not stay neatly inside OpenAI's walls. If a user connects records through one company, asks questions through another and receives advice from a model supplied by a third, who owns the failure when the answer is dangerously wrong? The front end will point to disclaimers - and the infrastructure provider to consent. The model company will point to intended use. Guess who actually ends up in the hospital bed.
A patient will point to the hospital bed.
For founders building symptom checkers, medical-record copilots or health dashboards on top of large language models, this case is a cleaner warning than most guidance documents. Do not market trust and then bury the risk in terms of service. Do not let a chatbot act like a clinician in the parts of the conversation where a real clinician would send the user to urgent care. Frankly, if your product cannot recognize the moment to stop answering and escalate, it is not ready for consumer health.
Winters' case may fail. Complaints are allegations, not findings - and OpenAI has not been found liable for any of them. Neither Sam Altman nor OpenAI has commented on the specific allegations in the reports reviewed.
Still, the lawsuit has already done something useful. It has stripped the issue down to conduct, not branding. If an AI system gives a diagnosis, recommends immobility and uses a man's faith to keep him calm while blood clots are forming, the label on the product is not the only thing a court should care about.
Also read: Small businesses are using AI to grow without hiring while big tech fires its way to the future • Genesis AI is in talks to raise $500 million at a $3 billion valuation just a year after emerging from stealth • Linus Torvalds tells AI critics to fork Linux or walk away and the software industry should listen
