OpenAI rolls out ChatGPT Health to all US users, claiming its models can out-reason clinicians

OpenAI is pushing ChatGPT Health out to every adult in the US on Thursday, betting that its AI can handle sensitive medical questions better than a doctor. Ashley Alexander, OpenAI’s vice president of health product, made the sharp-edged claim during a briefing that the company’s models “are now capable of reasoning at levels that are better than clinician level.” It’s a bold line—and one that the company immediately walked back when pressed for evidence. Health lead Karan Singhal tempered the claim, pointing only to “individual studies” like a Harvard and Stanford paper that point in that direction. The careful backpedal is notable, especially given the timing: a Florida pastor sued OpenAI just a day earlier, alleging ChatGPT gave him “extremely dangerous medical recommendations” that led him to delay treatment for a pulmonary embolism.

The broader launch runs on GPT-5.6 Sol, which OpenAI calls its strongest model yet for health. The key user-facing shift is removing friction. Instead of navigating to a dedicated ChatGPT Health tab, users can now ask questions about their connected records in the main chat window. The system will ask for permission before pulling personal health data to tailor a response, and OpenAI stresses that all conversations are encrypted both at rest and in transit—with health data getting an extra layer of protection. Users retain control over what they connect, from post-visit notes and lab results to medications, pulling data from providers and consumer apps like Apple Health, Weight Watchers, and MyFitnessPal.

The rollout lands during a trust crisis for AI in medicine, where the gap between marketing and peer-reviewed reality can have real consequences. OpenAI’s own disclaimer says the feature “supports, not replaces, professional care.” That’s the legally required safety net. But telling the public an AI reasons at a level beyond a clinician—then admitting that’s not quite right—creates a whiplash effect that doesn’t inspire confidence. It’s the classic Silicon Valley move: launch with a world-changing claim, then quietly add the asterisks.

For now, the tool is available to logged-in users 18 and older on web and iOS, across free, Go, Plus, and Pro plans. The real test won’t be in briefing-room claims or encryption specs. It’ll be in the first wave of users googling symptoms at 2 a.m. and deciding whether an AI’s answer is good enough to skip the doctor’s office.