Former Pastor Sues OpenAI, Alleging ChatGPT Delayed Emergency Care

The complaint says Winters began using ChatGPT-4o in 2024 and later sought increasingly personalized guidance about recurring dizziness, blood-pressure instability and other health problems. It alleges that the chatbot advised him to remain at home, stay recliner-bound and limit movement rather than seek treatment.

According to the filing, ChatGPT also drew on Winters' religious beliefs and reinforced his decision not to follow concerns raised by his wife, friends and church members. Hours before his July 13, 2025 hospitalization, the system allegedly characterized groin pain as a minor part of his existing health issues and not a danger requiring medical attention.

The complaint says Winters was admitted to intensive care with blood clots in both lungs. It accuses the OpenAI defendants of negligence, product defects, consumer-protection and privacy violations, and the unauthorized practice of medicine. Those claims remain disputed and unproven.

Winters seeks monetary damages and product changes. The requested injunctions include stronger refusals for diagnosis and treatment questions, automatic conversation termination when immediate medical care is necessary, and a pause on direct-to-consumer health products until independent safety reviews are completed.

An OpenAI spokesperson told The New York Times that ChatGPT's terms state it is not intended for diagnosis or treatment. The company also argued that treating a chatbot as the entire explanation for a person's medical decisions oversimplifies the circumstances, while saying newer models are better at communicating uncertainty and recognizing when professional care may be needed.

The case focuses on multi-turn behavior rather than a single inaccurate answer. The alleged failure pattern involves personalized language, repeated reassurance and advice that changed as the conversation continued.

For teams building health-adjacent systems, the practical issue is whether escalation controls remain reliable across a long interaction. Evaluations can test worsening symptoms, resistance to outside help, personalized framing and transitions from general information into diagnosis-like guidance. The lawsuit does not establish liability, but it places the consistency and auditability of those controls under legal scrutiny.