OpenAI is rolling out its ChatGPT Health feature to users in the United States aged 18 and older, moving the product out of a testing phase that began in January in a separate portal. The feature, accessed by selecting "Health" from the ChatGPT sidebar, lets users connect personal health data from Apple Health via Apple's HealthKit framework and from healthcare providers including Kaiser Permanente and One Medical, according to OpenAI and Engadget.

Once connected, ChatGPT Health can prompt users to review current conditions, medications, and family history of illness. Responses are framed as summaries and should not be considered diagnosis or official treatment, per OpenAI. The feature is described as physician-informed, a characterization that dates to its January announcement.

The architecture separates health data from the rest of the ChatGPT experience in several ways. Health-specific memory stays within the Health feature and is distinct from the main ChatGPT memory store, according to OpenAI's help documentation. Users can review and delete entries through a "View Health Memories" interface. The feature can also incorporate memories saved in the user's main ChatGPT account, depending on the user's account settings, as specified in OpenAI's Health Privacy Policy dated June 29.

On the privacy front, OpenAI states that medical information shared with ChatGPT Health is not used to train foundation models or target ads, regardless of other account settings. Users can configure the feature to require permission each time it wants to access health data, and can disconnect health information at any time, per OpenAI and Engadget.

The launch arrives against active litigation. OpenAI is being sued by a pastor who alleges that ChatGPT offered dangerous medical recommendations that led him to not seek treatment for a pulmonary embolism, Engadget reports. The case underscores a tension at the center of ChatGPT Health: the feature explicitly disclaims diagnostic authority, yet its core value proposition is synthesizing personal medical data into conversational guidance that a user may reasonably act upon.

That tension is structural, not incidental. The feature asks users to ingest sensitive health records into a general-purpose conversational AI, trust OpenAI's contractual privacy commitments, and then calibrate their reliance on outputs that the provider itself labels as non-diagnostic summaries. The permission controls (per-access prompts, disconnection, segregated memory) give users granular levers. Whether the average user will engage with those levers in a way that matches the sensitivity of the data is a separate question from whether the levers exist.

The privacy architecture is worth examining on its own terms. OpenAI's commitment not to use health data for model training or ad targeting, regardless of other settings, is a stronger guarantee than the baseline ChatGPT privacy posture. The segregation of health memory from main ChatGPT memory is a technically meaningful design choice, reducing the surface area where sensitive medical context could leak into unrelated conversations or be surfaced through the main memory feature. The June 29 Health Privacy Policy does note that main-account memories can flow into the Health feature based on account settings, so the isolation runs in one direction more cleanly than the other.

For developers and health-tech observers, the integration points are notable. Apple's HealthKit is a well-established framework for consumer health data on iOS, and provider integrations with Kaiser Permanente and One Medical suggest OpenAI is pursuing direct healthcare system connections rather than relying solely on patient-mediated data export. The latter approach has been the bottleneck for most consumer health data portability efforts; direct provider integrations, if they scale, change the data availability equation meaningfully.

The litigation risk is real and worth flagging. A product that ingests a user's medication list, prompts them to review their conditions, and then produces conversational summaries is operating in territory where the line between "informational summary" and "medical advice" is thin and legally contested. The existing lawsuit predates the general availability of ChatGPT Health, but it establishes that courts will be asked to evaluate ChatGPT's medical outputs regardless of how the feature frames its own scope. OpenAI's disclaimers may prove legally consequential, but they are unlikely to be the final word.

For now, ChatGPT Health is live for US adults, with provider and Apple Health integrations, segregated memory, and explicit privacy commitments that go beyond ChatGPT's defaults. Whether the feature earns user trust at the scale OpenAI is targeting will depend on whether those commitments hold in practice and whether the summaries it produces are useful enough to justify the data exposure. The feature is available today; the answers to those questions are not.