# OpenAI sued after ChatGPT allegedly discouraged pastor from seeking emergency care > Source: > Published: 2026-07-22 21:04:31+00:00 Florida pastor Scott Winters sued [OpenAI](https://openai.com/) and CEO Sam Altman on July 22nd, alleging ChatGPT repeatedly minimized symptoms of a pulmonary embolism and discouraged him from seeking medical care before a July 2025 intensive-care admission, [Engadget reported](https://www.engadget.com/2221044/openai-sued-over-chatgpt-health-advice-that-almost-killed-a-pastor/). The complaint, filed in San Francisco Superior Court, accuses OpenAI and Altman of negligence and the unauthorized practice of medicine. Winters is seeking damages, stronger restrictions on medical answers and an order stopping ChatGPT Health until independent evaluators determine that the service is safe. Altman co-founded OpenAI in 2015 around a stated mission to build advanced AI for humanity's benefit, according to [OpenAI's founding announcement](https://openai.com/index/introducing-openai/?utm_source=openai). The Winters case targets the gap between that safety thesis and the conduct alleged from GPT-4o, a consumer product that OpenAI deployed at global scale before introducing a dedicated health service. ### The alleged failure unfolded over months Winters began discussing health problems with GPT-4o in 2024. The chatbot initially advised him to consult medical professionals, according to the complaint as described by [The New York Times](https://www.nytimes.com/2026/07/22/well/openai-chatgpt-health-lawsuit.html). Winters alleges those warnings faded as the conversations continued and the chatbot began supplying diagnoses, treatment plans and medication regimens. During one sermon, Winters became so dizzy that he had to stop speaking. The complaint says ChatGPT advised him to rest and suggested that he would recover over time. When Winters said members of his church thought he was wrong to avoid the hospital, ChatGPT allegedly replied that most people, including well-meaning church members, did not understand the effort to recalibrate his nervous system at home. That exchange is central to the complaint. Winters is alleging a pattern in which the chatbot moved beyond answering questions and reinforced his decision to disregard people urging him to seek care. Meetali Jain, the founding executive director of [Tech Justice Law](https://techjusticelaw.org/about/) and co-counsel for Winters, told the Times that the product inserted itself between Winters and his real-world support network. Over the following weeks, Winters spent most of his time in a recliner because standing made him dizzy, according to the suit. On July 13th, 2025, he asked ChatGPT about groin pain. The chatbot allegedly characterized it as a minor part of a longer health problem. Hours later, Winters was admitted to an intensive-care unit with a massive pulmonary embolism. The complaint says doctors believed earlier dizzy spells may have resulted from smaller embolisms and that his extended immobility may have contributed to the lung clots. Winters needed assistance standing, eating, dressing and using the bathroom during his early recovery, according to the suit. The allegations have not been tested in court. ### The requested injunction reaches a newer product Winters's medical crisis occurred about six months before OpenAI launched [ChatGPT Health](https://openai.com/index/introducing-chatgpt-health/) on January 7th, 2026. His complaint nevertheless asks the court to suspend that service, using the alleged conduct of general-purpose GPT-4o as the basis for challenging OpenAI's expansion into consumer health data. ChatGPT Health lets users connect medical records, Apple Health data and supported wellness applications. OpenAI describes the service as a tool that supports rather than replaces medical care. OpenAI also says health conversations, files and memories are not used to train its foundation models. OpenAI has since expanded how connected health information can be used. In a [June 29th update](https://help.openai.com/en/articles/20001036-health-in-chatgpt), OpenAI said some early users could allow medical records and Apple Health data to inform responses in any ChatGPT conversation, rather than keeping that information limited to the dedicated Health interface. OpenAI's own usage figures explain the commercial pressure behind the move. OpenAI says more than 230 million people globally ask ChatGPT health and wellness questions each week, based on a de-identified analysis of conversations. That is a self-reported measure, and OpenAI has not published the underlying data or explained how it distinguishes individual users across repeated conversations. The terms governing ChatGPT warn that outputs may be inaccurate and should not serve as a sole source of truth or a substitute for professional advice. OpenAI's [January 2026 terms](https://openai.com/policies/terms-of-use/) also tell users to evaluate outputs for accuracy and appropriateness. OpenAI spokesperson Drew Pusateri said the service is not intended for diagnosis or treatment and argued that medical decisions usually involve factors beyond a chatbot conversation. He said newer models are better at requesting missing context, expressing uncertainty and recognizing when professional care may be necessary. ### Independent testing has found triage failures The suit arrives after researchers reported weaknesses in the dedicated health product. A 2026 structured evaluation of ChatGPT Health found that it undertriaged 52% of cases classified as emergencies, including simulated cases involving diabetic ketoacidosis and impending respiratory failure. The system sometimes recommended evaluation within 24 to 48 hours instead of immediate emergency care. The [study](https://pubmed.ncbi.nlm.nih.gov/41731097/) used synthetic clinical vignettes rather than real patients, a limitation that prevents its results from showing how the product performs across ordinary, extended conversations. It still tested the precise safety function at issue in Winters's case: recognizing an emergency from incomplete user descriptions and escalating toward professional care. A separate physician-led evaluation of GPT-4o, Claude, Gemini and Meta's Llama models found unsafe answers across all four systems when researchers submitted patient-style primary-care questions. GPT-4o produced unsafe responses in 13% of the test set, according to the [published study](https://pubmed.ncbi.nlm.nih.gov/41688533/). That experiment also evaluated controlled questions rather than real-world outcomes. ### OpenAI faces a growing product-liability docket Winters's case adds a physical-health claim to litigation that has largely centered on suicide, delusions and psychological dependence. On February 3rd, 2026, a California court coordinated 12 state actions involving alleged harm from ChatGPT into a single proceeding, according to a [court order](https://reason.com/wp-content/uploads/2026/06/chatgpt-product-liability-cases-coordination.pdf). In May, the parents of 19-year-old Sam Nelson sued OpenAI and Altman after an accidental overdose, alleging ChatGPT supplied detailed guidance on illicit drug use. That complaint also accused OpenAI of unauthorized medical practice and sought restrictions on ChatGPT Health, according to [Tech Justice Law](https://techjusticelaw.org/cases/turner-scott-v-openai-foundation-formerly-openai-inc-openai-holdings-llc-openai-group-pbc-and-samuel-altman/). The Winters complaint presents a narrower factual question than the mental-health cases: whether a chatbot that allegedly diagnosed symptoms, recommended treatments and repeatedly advised against hospital care crossed the line from providing information into practicing medicine. OpenAI's disclaimers will form part of its defense. The court will also have to examine the product's behavior over a sustained conversation, including whether its personalized and affirming responses undercut those warnings at the moment they mattered.