ChatGPT is starting to pull US users' medical records, lab results and Apple Health data into the same chat surface.
OpenAI said on July 23 that Health in ChatGPT is rolling out to signed-in users in the United States aged 18 and older. The first launch covers the web and iOS, and is available across Free, Go, Plus and Pro plans. Users can choose to connect Apple Health, supported US hospital records, One Medical or Function Health. Once authorized, ChatGPT can bring lab results, medications, recent visits, sleep and activity data into a single conversation.
This is more than a new health tab. OpenAI's starting point is that more than 300 million people already ask ChatGPT health-related questions each week, from understanding lab reports to preparing follow-up visits and parsing what a doctor said. The change is that the model no longer has to rely only on what users retell; with permission, it can read structured personal health data.
The entry point moves back into ordinary chat
OpenAI had already tested a standalone Health area earlier in the year. The redesign is explained by one early usage signal: more than 70% of health-related conversations among early users happened outside the dedicated health experience. In ordinary life, allergies, injuries, sleep or medication come up while planning a meal, a workout, a trip or a weekend activity, and users rarely want to switch spaces first.
Health therefore remains in the sidebar as a profile and data-management center, while the information can be used in normal chats. If a user asks where to take a child on the weekend, ChatGPT can, after permission, account for a recent injury. If the question is about dinner, it can factor in diet restrictions or activity synced from Apple Health.
OpenAI quoted an early user, Blake, as saying:
"The most useful part has been turning scattered medical history into something I can actually understand and explain."
In practice, that means turning diagnoses, imaging notes and surgery records into a timeline a patient can explain to a physical therapist or coach. The value is real, and so is the danger: if the model misreads an old medication, an old condition or a wearable metric, its advice may look more grounded than it is.
Privacy commitments sit in the front row
OpenAI put detailed privacy language near the top of its announcement. All ChatGPT conversations are encrypted in transit and at rest, and connected medical records and Apple Health data receive additional encryption. Those records, and the conversations that use them, will not be used to train foundation models or for ad targeting. After a user disconnects an account, synced data is deleted from OpenAI systems within 30 days.
Permissions are layered as well. By default, ChatGPT asks before it uses Health information each time. Users can allow one-time access or keep access on. If ChatGPT is about to send content that may include health information through another connector, the system asks for an additional confirmation.
The design is aimed at the trust barrier. Health data is more sensitive than calendars, email or shopping history. For OpenAI, turning ChatGPT into a personal health entry point is not only a question of whether the model can read a lab result. It is also a question of whether users believe the company will not turn medical records into training fuel.
Medical capability moves into the spotlight
OpenAI framed model capability in two ways: GPT-5.5 Instant brings frontier health intelligence to free users, while the paid GPT-5.6 Sol model is its strongest system for health tasks. The company says every GPT-5.6 model beats GPT-5.5 on HealthBench Professional, an evaluation designed with hundreds of doctors worldwide and covering accuracy, safety, communication, context awareness, completeness and escalation to care.
The Verge put the risk near the top of its coverage: OpenAI is making large claims about medical reasoning, but the company also acknowledges that ChatGPT can make mistakes and cannot replace qualified medical professionals. People reported the same day on a medical-advice lawsuit against OpenAI, in which the plaintiff alleges ChatGPT guidance delayed care. Whatever the result, the case tightens the external environment around a broad Health rollout.
The real test for this category is graceful escalation and refusal. The more believable the product becomes, the more clearly the interface must surface warnings such as "contact a doctor soon," "this reading may be affected by device error" or "check the original record," instead of wrapping complex histories in a smooth answer.
Three questions will matter first in China
The first is data access. US users can connect Apple Health and parts of their medical record. In China, common health data is spread across hospital mini programs, insurance platforms, exam providers and wearable apps, making interoperability harder.
The second is responsibility. OpenAI repeatedly says Health only helps users understand information and prepare questions, and does not replace professional care. If similar products enter Chinese-language settings, the platform, hospital, doctor and model provider will need clearer responsibility boundaries than ordinary chat products require.
The third is the business boundary. Health spans free plans through Pro, suggesting OpenAI wants healthcare to become a basic ChatGPT use case rather than a product sold only to hospitals or insurers. The larger the entry point, the larger the cost of mistakes, privacy failure and weak explainability.
ChatGPT Health's wider release pushes AI healthcare from "the user asks a question" toward "the model answers with personal records in hand." The convenience comes from the same single entry point as the risk. The next checkpoints are concrete: more medical-record sources, non-iOS support, whether public HealthBench details can support external clinical review, and whether high-risk advice can withstand independent audits.
Sources: OpenAI official announcement, The Verge, CocoLoop, TechCrunch, Medical Economics and People; checked US availability, age limit, plan and platform coverage, 300M weekly health questions, consent mechanism, training and ad exclusions, 30-day deletion language and the medical-dispute context.