OpenAI is rolling out Health in ChatGPT to logged-in U.S. users 18 and older, expanding a January health experiment into the main ChatGPT experience on web and iOS. The new version lets users connect Apple Health and supported medical records, then allow ChatGPT to use that context when answering health-related questions.
The feature is aimed at a common problem in digital health: personal health information is scattered across patient portals, lab reports, wearable apps, visit notes, medication lists, and memory. OpenAI says more than 300 million people ask ChatGPT health-related questions each week, and the new Health tab is meant to make those conversations less dependent on manual uploads or half-remembered details.
That makes Health in ChatGPT potentially useful, but it also moves ChatGPT closer to some of the most sensitive information a consumer can share with a software platform. The product can help users prepare for appointments, compare lab results over time, summarize changes since a previous visit, or ask how sleep, activity, workouts, medications, and symptoms might fit together. It is not presented as a diagnostic tool, and OpenAI repeatedly says it is designed to support, not replace, professional medical care.
What Changed in the Rollout
The most important change is that Health is no longer only a separate place inside ChatGPT. OpenAI says early testing showed more than 70% of health-related conversations happened outside the dedicated health area, such as when a user was planning meals and wanted the system to account for allergies or dietary restrictions. In the new rollout, ChatGPT can ask for permission to use connected health information inside ordinary conversations.
Users can connect Apple Health, supported medical records from U.S. hospital systems, One Medical, and Function Health, according to OpenAI’s Help Center documentation. Apple Health requires an iPhone. After syncing, the Health tab can show recent data and trends, suggested prompts, synced records, connected accounts, and past health conversations.
The company’s examples are deliberately practical. A user might ask ChatGPT to explain a visit note in plain language, identify what changed in a lab result, help prepare questions for a follow-up appointment, or consider a recent injury while planning lower-impact weekend activity. The product can also use context such as medications, post-visit notes, care-team notes, lab results, activity, and sleep if the user has connected that data and allowed access.
The Privacy Controls Are Central
OpenAI says connected medical records and Apple Health information, along with conversations that use that information, are not used to train its foundation models or target ads. The company also says all ChatGPT conversations are encrypted at rest and in transit, while information connected in Health gets additional encryption protections.
The permission model matters because Health information can now be used outside the Health tab. By default, ChatGPT asks before using connected medical records or Apple Health information to personalize a response. Users can approve a single request, always allow access, change the setting later, or explicitly add @Health to a message when they want connected health context included.
There are still practical limits readers should understand before connecting data. OpenAI says users can disconnect accounts from Health, and synced data from that source is deleted from OpenAI’s systems within 30 days. But information already included in ChatGPT conversation history remains until those conversations are deleted. Connected records may also be incomplete or out of date, so OpenAI advises users to update current conditions, medications, and family history, and to check important details against the original source.
Safety Claims Meet a Harder Real-World Test
OpenAI is pairing the rollout with stronger performance claims. The company says GPT-5.5 Instant improved health responses for free users, while GPT-5.6 Sol is its strongest model for health on more complex questions. Its release note says Health can help users view lab results, medications, activity, sleep, and other health information in one place, while preparing for appointments and thinking through questions to discuss with a healthcare professional.
The harder issue is not whether health-focused models can produce better explanations than older chatbots. It is whether they reliably fail safely when a user describes dangerous symptoms, minimizes risk, withholds context, or wants reassurance instead of escalation. Health advice is a high-trust interaction, and the most damaging failure is often not an absurd answer. It is a plausible, confident, personalized answer at the wrong moment.
Independent research has already raised that concern. A Nature Medicine study stress-tested ChatGPT Health triage recommendations using 960 responses across clinician-authored vignettes. The researchers found the system undertriaged 52% of gold-standard emergencies, including cases where diabetic ketoacidosis or impending respiratory failure were directed toward delayed evaluation rather than emergency care. The same study found crisis-intervention messages appeared inconsistently across suicidal-ideation scenarios.
OpenAI says physicians extensively tested Health in ChatGPT before release and that the system can still make mistakes. That caveat is not boilerplate here. The product is entering a consumer environment where people may ask about symptoms late at night, after frustrating appointments, without complete records, or while already anxious. A better health assistant can help people ask sharper questions of a doctor; a bad one can become a substitute authority if the interface feels too personal or certain.
The Lawsuit Adds Pressure
The national rollout also arrived one day after a lawsuit filed in San Francisco County Superior Court accused OpenAI of giving dangerous medical advice through ChatGPT-4o. According to SFGATE’s report, Florida pastor Scott Winters alleges that ChatGPT discouraged him from seeking medical care for dizziness and other symptoms before he was hospitalized with multiple pulmonary blood clots. The allegations have not been proven in court.
The case is separate from the new Health rollout, which runs on later models and includes product-specific safeguards. But the timing underscores why the product’s design details matter. A health chatbot that can read records, remember context, and respond in a familiar tone may feel more credible than a search result or symptom checker. That credibility is useful when it helps someone prepare for care. It is dangerous if a user treats the answer as a clinician’s judgment.
How Readers Should Use It
Health in ChatGPT is best treated as a health-information organizer and appointment-prep tool, not as a triage line. The safest use cases are summarizing records, translating medical terminology into plain language, comparing a new result with older data, building a timeline, preparing questions, and identifying which source document should be checked again.
Users should be more cautious when asking whether symptoms require urgent care, whether to delay treatment, whether to stop medication, or whether a doctor’s recommendation is unnecessary. Those are exactly the moments when a chatbot’s confident wording can cause harm. If symptoms are severe, sudden, worsening, or difficult to interpret, the right comparison is not ChatGPT versus Google. It is ChatGPT versus a qualified clinician, nurse line, urgent care, emergency department, or the instructions already provided by a care team.
The product is also a reminder that AI privacy choices are becoming more concrete. Connecting a health account is not the same as typing a one-off question. It creates a standing relationship between a chatbot and a live body of sensitive data. Before turning it on, users should review which sources are connected, whether permission prompts remain enabled, what records are current, and how comfortable they are with health context appearing in ordinary ChatGPT conversations.
OpenAI’s broader bet is that people already ask ChatGPT health questions, so the safer path is to ground answers in better context and stronger controls. That may be right. But the trust test is no longer just whether ChatGPT can explain a lab result. It is whether the product can keep users oriented toward real medical care when the answer matters most.
