OpenAI launches Health in ChatGPT, letting US users connect medical records and Apple Health

OpenAI has begun rolling out Health in ChatGPT to logged-in US users 18 and older on web and iOS, across the Free, Go, Plus, and Pro plans. The feature lets people securely connect Apple Health and supported medical records so ChatGPT can reference that information across ordinary conversations, not just inside a dedicated health tab.

What's new

Users can connect Apple Health along with medical records from supported US hospital systems, One Medical, and Function Health. Once connected, ChatGPT can draw on details like medications, lab results, recent visits, sleep, and activity data to compare a new test result against prior ones, summarize what has changed since a last appointment, or factor a dietary restriction or recent injury into everyday suggestions. By default, ChatGPT asks permission before using this connected data in a given response; users can approve requests individually or turn on always-allow access, and can disconnect any account at any time, with synced data deleted from OpenAI's systems within 30 days.

OpenAI says the underlying models have been tuned specifically for health conversations. GPT-5.5 Instant, available to free users, and GPT-5.6 Sol, available to paid users, were developed with input from physicians who built realistic health scenarios and rubrics scoring accuracy, safety, communication, context awareness, completeness, and appropriate escalation to professional care. Every GPT-5.6 model reportedly outperformed GPT-5.5 on HealthBench Professional, OpenAI's internal evaluation for difficult health tasks. OpenAI states connected medical records and Apple Health data, and any conversations that use them, are not used to train its foundation models or to target ads, regardless of a user's general model-training setting, and that memories are not created directly from that connected data.

Context

OpenAI first introduced a limited, separate health experience earlier this year that required users to enter a dedicated space to get the full benefit. According to the company, more than 70% of health-related conversations from that early group happened outside the dedicated space entirely — people asked health-adjacent questions while doing ordinary things like planning meals, and the extra step of switching contexts got in the way. Health in ChatGPT is the response to that feedback: the connected data now travels with the user across ChatGPT rather than being walled off. As OpenAI put it in its announcement, "We're continuing to improve the models behind health conversations so ChatGPT can reason more carefully across complex details, explain health information clearly, ask for missing context, and recognize when professional care may be needed."

Why it matters

Health in ChatGPT pushes a general-purpose chatbot deeper into a domain — personal medical data — where the cost of a wrong or overconfident answer is unusually high, and where privacy expectations are correspondingly strict. OpenAI's own framing leans on scale to justify the move: "Every week, more than 300 million people turn to ChatGPT with health-related questions." Whether that volume of usage was previously well served is an open question the company is now trying to answer by attaching structured, longitudinal personal data to those conversations rather than relying on one-off descriptions typed in by users.

The privacy architecture — permission gates before each use, a hard line against training on connected health data, and red-team testing aimed specifically at cross-plugin information leaks (like a running plan built from Apple Health data getting shared with the wrong contact) — signals OpenAI anticipates scrutiny from regulators and health-privacy advocates alike. It also puts pressure on Google and Anthropic, both of which have positioned their assistants for health-adjacent use cases without shipping anything as directly tied to real medical records and wearable data at this scale. If Health in ChatGPT works as described and avoids a high-profile privacy or safety incident, it becomes a template other consumer AI products will likely copy; if it stumbles, it becomes the cautionary case regulators cite.