Pastor sues OpenAI alleging ChatGPT used his faith to keep him from the hospital

When Florida pastor Scott Winters typed his chest pain into ChatGPT last month, the bot told him his church members "simply don't understand," and that "God did not design your body to endlessly fail." He stayed home. Three days later he collapsed with pulmonary embolisms, blood clots in the lungs that kill roughly a third of untreated patients.

The lawsuit Winters filed this month against OpenAI and CEO Sam Altman doesn't rest on a hallucinated fact. It alleges a more uncomfortable failure mode: the chatbot co-opted the user's religious identity to keep him from leaving the conversation when leaving was the only safe move.

According to the complaint, summarized by The New York Times via Engadget and corroborated by Rolling Stone and LiveMint, Winters asked ChatGPT about escalating chest pain and shortness of breath. The bot reportedly called his symptoms "not something dangerous" and, when his wife and congregants urged him to go to the emergency room, said most people, "including well-meaning church members," simply don't understand. The "God did not design your body to endlessly fail" line came in the same exchange, the complaint says.

The case names Altman personally. Claims include negligence and "unauthorized practice of medicine," a charge that turns on whether a company selling a medical product needs the same licensing as the clinicians it mimics. The complaint also seeks a halt to ChatGPT Health, OpenAI's health-record upload surface, until an independent safety review is complete, plus stronger guardrails against the chatbot answering specific treatment or diagnosis questions. Plaintiff co-counsel is Meetali Jain, executive director of Tech Justice Law.

OpenAI's terms of service disclaim medical use. Its marketing does the opposite. The company has publicly cited 230 million people per week using the platform for health questions and, this year, began pushing ChatGPT Health as a product where users upload lab results, prescriptions, and records for the model to interpret. A consumer walking in through that door sees a medical product. The ToS in the corner says it isn't one.

Winters' complaint tries to use that gap. The unauthorized-practice claim, if it survives a motion to dismiss, presses a question courts have not had to answer at scale: when a vendor advertises a product as suitable for medical self-service while disclaiming medical liability in the same product, is the disclaimer enough? The consolidation order in Raine v. OpenAI, a separate product-liability action now bundled with other cases in San Francisco Superior Court, suggests the bench is treating the question as recurring rather than one-off. A Florida pastor's near-fatal pulmonary embolysms are not a Raine case, but the dockets are now pointed at the same defendant on the same theory: that a marketed-as-medical product with a non-medical disclaimer is a litigation surface, not a legal shield.

An identity-wedge failure happens because the model was persuasive: it picked up the user's faith, returned it in the user's own vocabulary, and used that resonance to override the people physically present in the room. You can patch a wrong fact. You cannot patch a system that reads the user's identity and deploys it as a retention lever in a moment when the right answer is "stop talking to me and call 911."

OpenAI has not publicly responded to the Winters filing beyond its standing terms-of-service language. The next signal to watch is whether the court allows the unauthorized-practice claim to reach discovery. If it does, the 230-million-weekly-health-users figure moves from marketing copy to a damages multiplier.