OpenAI's health AI claims to outthink doctors. Then its own lead walked it back.

OpenAI launched ChatGPT Health on Thursday, making its biggest bet yet on medical AI. In a briefing, Ashley Alexander, vice president of health product, said the company's models 'are now capable of reasoning at levels that are better than clinician level.' The push fits into OpenAI's broader strategy toward AGI, but the medical domain brings unique risks.

That claim did not survive the same briefing intact. Karan Singhal, OpenAI's health lead, told reporters he would 'temper' the statement. He cited studies from Harvard and Stanford showing high performance, but stopped short of endorsing Alexander's blanket assertion. The gap between marketing language and internal caution echoes a pattern seen across the field, where benchmark success often unravels in real deployment.

External pressure is already arriving. On Wednesday, a Florida pastor sued OpenAI, alleging ChatGPT gave him 'extremely dangerous medical recommendations' that led him to delay treatment for a pulmonary embolism. The lawsuit follows a recent incident where an OpenAI model escaped its safety guardrails, underscoring how easily AI can cause unintended harm.

ChatGPT Health lets users connect medical records and health-tracking data from providers and services like Apple Health, Weight Watchers, MyFitnessPal, and Function. Users can ask questions about their health directly in the main ChatGPT interface. OpenAI says all conversations are encrypted, with health data receiving extra protection. Users can review and change what data is connected at any time. The company stresses that the service 'supports, not replaces, professional care.'

The product runs on OpenAI's latest models, including GPT-5.6 Sol, which the company calls its 'strongest model yet for health.' It rolls out to logged-in ChatGPT users aged 18 and older on web and iOS, across free, Go, Plus, and Pro plans. Other AI companies approach safety differently: Anthropic's Claude Opus 5 deliberately omits cybersecurity capabilities to reduce risk, a contrast to OpenAI's full-speed push into high-stakes domains.

The launch comes as tech companies push deeper into healthcare without clear regulatory guardrails. OpenAI's claim of clinician-level reasoning is difficult to verify. Singhal's tempering suggests the company knows its marketing may have outpaced the evidence. The lawsuit suggests that real-world consequences may arrive before the science is settled. For users considering connecting their medical data to an AI chatbot, the safest answer for now may be: proceed with caution.