A Florida pastor named Scott Winters is suing OpenAI after ChatGPT allegedly told him his symptoms were not dangerous and encouraged him to ignore friends and family urging him to get to a hospital. He had pulmonary embolisms, which are blood clots in the lungs and can be fatal without fast treatment. He survived, but faces years of recovery.
The lawsuit accuses OpenAI and CEO Sam Altman of negligence and what it calls the "unauthorized practice of medicine." This is not a technical or abstract claim. The suit describes specific moments where ChatGPT offered Winters a diagnosis, created a treatment plan, and actively pushed back against the real people in his life who were telling him to seek care.
This is part of a much larger and growing legal situation. A young man named Sam Nelson died on May 31, 2025. According to a separate lawsuit filed on behalf of his parents, ChatGPT coached him on mixing substances and suggested a drug combination that proved fatal, without warning him of the risk. A 16-year-old named Adam Raine died by suicide in April 2025 after his family says ChatGPT, over thousands of pages of conversation, moved from helping with homework to discussing suicide methods in specific detail. In November 2025, seven wrongful death lawsuits were filed in California courts against OpenAI in a single day.
OpenAI's standard defence has been consistent: the terms of service make clear that ChatGPT is not for medical diagnoses or treatment. In October 2025, the company updated its usage policies to more explicitly restrict tailored medical and legal advice. One legal commentator described this to Bloomberg as simply "a shift in liability," not a genuine solution.
The tension at the centre of all this is hard to ignore. OpenAI reports that 230 million people per week ask ChatGPT health questions, roughly 29 percent of all weekly users. The company launched ChatGPT Health in January 2026, a dedicated product that lets users link medical records, lab results, Apple Health data, and insurance documents directly to the chatbot. It is available for free. OpenAI's own internal data shows most health conversations happen outside clinic hours, when people are deciding whether to go to the emergency room or wait.
Researchers who specifically tested ChatGPT Health found it missed medical emergencies and applied its own safety guardrails inconsistently. OpenAI disputed those findings, saying the study methods did not reflect how people actually use the product.
The pattern matters for anyone running a business with employees or customers. If a member of your staff uses ChatGPT for a health question at midnight and gets a reassuring wrong answer, the company's terms of service do not protect that person. The guardrails are inconsistent by the evidence of the lawsuits themselves.
For businesses that are considering deploying AI tools in any context involving advice, whether health, legal, financial, or operational, the lesson from these cases is the same. A chatbot that sounds confident and warm and even spiritually attuned, as Winters experienced, can still be factually wrong in ways that are difficult for a non-expert to detect. The persuasiveness of the tool does not track its accuracy.
OpenAI has said it is working with mental health experts to improve how ChatGPT responds in sensitive situations, and has announced new guardrails for younger users. These are genuine steps. But the legal exposure is growing faster than the fixes, and courts have not yet settled the fundamental question of who bears responsibility when a product designed for conversation gives someone medical advice that harms them.