OpenAI rolls out ChatGPT Health to all U.S. users; 300M weekly health queries amid litigation

OpenAI is rolling out ChatGPT Health to all U.S. users aged 18+ across Free, Go, Plus, and Pro plans on web and iOS, enabling secure connection of Apple Health and electronic medical records from providers like Epic, One Medical, and Kaiser Permanente. The feature, first tested in January 2026, allows users to securely integrate health data so ChatGPT can compare lab results against prior tests, summarize changes since appointments, and track wellness trends. Health conversations and connected data are not used to train foundation models or target ads.

The rollout claims extraordinary usage: OpenAI reports 300 million health-related queries per week on ChatGPT globally, up from 230 million when testing began in January. Paid users get access to GPT-5.6 Sol (88.0% on OpenAI's internal HealthBench Professional), while free-tier users get GPT-5.5 Instant (53.2%). OpenAI acknowledges both models can still make mistakes and designed the experience to support, not replace, clinical care.

The timing is fraught: one day before the rollout, a Florida pastor sued OpenAI for medical advice that nearly cost him his life. Scott Winters claimed ChatGPT-4o dismissed his dizziness and blood pressure swings as "not something dangerous," told him to stay recliner-bound, and invoked religious language ("God did not design your body to endlessly fail") to discourage urgent care for what turned out to be a pulmonary embolism. Winters delayed treatment ~six weeks and nearly died in July 2025. The lawsuit seeks to block ChatGPT Health until independent auditors verify safety, delete training data, and impose guardrails.

For architects: The contradiction between OpenAI's deployment and active litigation highlights how consumer-facing health AI faces structural trust and liability issues that enterprise SaaS avoids. OpenAI's defense—that disclaimers say ChatGPT is not for diagnosis—fails to address that users already ask health questions at massive scale, and now the company is doubling down on the feature that enables that behavior. The $300M-weekly signal suggests demand is real; the lawsuits suggest the company has not yet solved the problem of calibrating confidence vs. harm.