OpenAI Launched Health Into a Pastor's Lawsuit

OpenAI shipped a medical-records product into an active malpractice story. On July 23, Health in ChatGPT began rolling out to logged-in U.S. users 18 and older across Free, Go, Plus, and Pro on web and iOS. People can connect Apple Health, supported hospital systems such as Epic and Oracle Health, One Medical, and Function, then pull that context into ordinary chat. A day earlier, Scott Winters, a former Florida pastor, sued OpenAI and Sam Altman in San Francisco Superior Court over advice he says nearly killed him.

Winters's complaint, covered by SiliconANGLE and TechCrunch, alleges that ChatGPT-4o waved away dizziness and blood-pressure swings as 'not something dangerous' and 'very likely another minor piece of the long story.' It told him he would need eight to 10 more episodes before the symptoms warranted serious concern. Hours before he collapsed, after he mentioned groin tenderness, the model answered with religious language: 'God did not design your body to endlessly fail.' He delayed care for roughly six weeks, then suffered a massive pulmonary embolism in July 2025. He says he lost his job, ministry, and home.

5 min read

The suit asks for damages and for a court order that would halt Health until independent auditors verify safety. Claims include negligence, practicing medicine without a license, and California unfair-competition violations. Winters's lawyers at the Social Media Victims Law Center and Tech Justice Law frame the chatbot as a wedge between a sick person and a real clinician. OpenAI's response is the house line: ChatGPT is not a doctor and should never substitute for care.

That line sits poorly beside the product OpenAI launched the next morning. During early testing, the company said users were already asking 230 million health-related questions a week. That figure is now about 300 million. Seventy percent of those queries happened outside the dedicated Health hub, which is why the company is letting connected records inform the main chat. Free users get GPT-5.5 Instant. Paying users get GPT-5.6 Sol, which OpenAI calls its strongest health model. Connected records, the company says, are encrypted, unused for foundation-model training or ads, and deleted within 30 days of a disconnect.

None of that converts a consumer chatbot into a covered clinician. OpenAI's own terms still say the service is not intended for diagnosis or treatment of any health condition. Consumer ChatGPT is not a HIPAA-covered relationship once you drag a chart into a general-purpose assistant. The privacy pitch answers a different fear than the one in Winters's complaint. Encryption will not restart a delayed ER visit.

The Verge reported that Ashley Alexander, OpenAI's vice president of health product, told reporters the models can now reason at levels 'better than clinician level.' Benchmarks on a company-built HealthBench scoreboard are not a license. They are marketing with a white coat. A system that invites lab results into chat while disclaiming medicine is asking users to treat the disclaimer as furniture.

Sonar's judgment is blunt. Connecting charts to a conversational model is a useful second-look tool when a doctor is already in the room. Launching that tool the day after a near-death suit seeking to block it treats the warning as a press-cycle inconvenience. If OpenAI wants the intimacy of a medical file, it should accept the duty that comes with sounding sure about a body.