ChatGPT Health: OpenAI's Apple Health and Records Rollout

On July 23, 2026, OpenAI began rolling out Health in ChatGPT to every logged-in US user aged 18 and older, on web and iOS, across the Free, Go, Plus, and Pro plans. The feature lets people securely connect Apple Health data and supported medical records, including from hospital systems Epic and Oracle Health, directly into the main chat window, so ChatGPT can answer questions with the person's own health history in view rather than in the abstract.

That is a significant expansion of scope for a consumer chatbot, and it arrives with a specific, sourced privacy pitch attached. It also arrives into a debate that predates this rollout by months: whether health data handed to a consumer AI product sits inside the same legal protections people assume cover anything called "medical," or outside them entirely.

OpenAI

Apple

Health in ChatGPT connects OpenAI's chatbot to Apple Health and hospital-system medical records.

July 23, 2026Health in ChatGPT wide US rollout date

18+Minimum age, US-only at launch

4Plans covered: Free, Go, Plus, Pro

~6 monthsSince the earlier limited pilot

What actually shipped

The mechanics are straightforward. A user opts in and links data sources: Apple Health for activity, sleep, and vitals; Epic or Oracle Health for hospital and clinic records; and, per OpenAI's rollout materials, additional platforms such as Function, MyFitnessPal, and One Medical. Once connected, that context is available inside ordinary chat conversations rather than confined to a separate health-specific tab, which was reportedly a friction point in the earlier version of the product.

How a connected record becomes a ChatGPT answer

Three categories of source data feed into the main chat interface, which sits behind stated limits on training, ads, and encryption.

Apple Health

Activity, sleep, vitals synced from iPhone and wearables

Epic & Oracle Health

Hospital and health-system electronic medical records

Function, MyFitnessPal, One Medical

Lab-testing, nutrition, and primary-care platforms

→

Health in ChatGPT

Contextualizes replies in the main chat window using connected data

↓

Not used to train models

Connected records and chats about them are excluded from foundation-model training

Not used to target ads

Health-connected data is walled off from ad-targeting systems

Extra encryption

Health-connected info gets additional encryption on top of standard at-rest and in-transit encryption

The safeguards apply to how OpenAI says it handles the data pipeline. They do not change ChatGPT's status as an informational tool rather than a HIPAA-covered medical provider.

The scale OpenAI has pointed to is large. Health-related queries on ChatGPT were already running at roughly 230 million a week during the narrower pilot, a figure OpenAI has said has since climbed to around 300 million weekly as health questions have become one of the more common categories of consumer use. Whatever else is true about the wisdom of the feature, the underlying demand for AI-assisted health information is not in question.

A quieter first attempt

This is not OpenAI's first pass at the idea. A more limited version, then called ChatGPT Health, launched to a narrower audience on January 7, 2026, built on health-data connectivity infrastructure from b.well and offering links to Apple Health, MyFitnessPal, Weight Watchers, and Function, among others. Reception was muted enough that the July relaunch reads as a deliberate second attempt: broader eligibility, the health-specific tab folded into the main interface, and a more explicit privacy framing attached to the announcement. Roughly six months separate the two launches, which is a short cycle for OpenAI to expand a feature this sensitive rather than shelve it.

A stethoscope photographed against a white background — A stethoscope, photographed as a stand-in for the distinction OpenAI is drawing: ChatGPT Health is positioned as an informational layer over a person's own data, not a diagnostic instrument or a replacement for a clinician. Photograph by Wikimedia user Stethoscopes, CC BY-SA 3.0.

The privacy commitments, and what they don't reach

OpenAI's stated safeguards are specific rather than generic. Connected medical records and Apple Health data, along with conversations that draw on them, are excluded from foundation-model training and from ad targeting. All ChatGPT conversations are encrypted at rest and in transit, and OpenAI says health-connected information receives an additional layer of encryption on top of that baseline. Those are real, falsifiable commitments, not marketing language, and they go further than what most consumer software volunteers about sensitive personal data.

What they do not do is change ChatGPT's regulatory category. HIPAA, the US federal law most people associate with medical privacy, governs "covered entities": hospitals, insurers, physician practices, and their business associates. A consumer chatbot is generally not one of those, so a medical record that a hospital must protect under HIPAA while it sits inside Epic is, once a user copies it into ChatGPT, no longer inside that legal perimeter. It becomes subject instead to OpenAI's own terms of service and privacy policy, which is a different and, critics argue, thinner form of protection with no dedicated federal health-privacy backstop filling the gap. Notably, OpenAI has a separate enterprise product, ChatGPT for Healthcare, built for hospitals and clinicians with a Business Associate Agreement available; Health in ChatGPT, the consumer feature covered here, is not that product and does not carry the same regulatory obligations.

Once a hospital record moves into a consumer chat window, it usually leaves the legal perimeter that required a hospital to protect it.
The HIPAA boundary consumer health AI sits outside of

Both readings of this are coherent. OpenAI can honestly say it has built encryption, training exclusion, and ad-targeting exclusion beyond what regulation requires of it. Privacy researchers can just as honestly say that voluntary safeguards are not the same as statutory ones, and that a policy commitment can change with a terms-of-service update in a way a law cannot. The rollout does not resolve that tension; it just puts more people's data inside it.

Informational, not diagnostic, in practice

OpenAI is explicit that ChatGPT Health is meant to help people understand their own information, not to diagnose or treat them, and the product surfaces that framing prominently. The distinction matters more than it might sound, because the line between "explaining a lab result" and "telling someone what to do about it" is not always sharp in a free-flowing conversation, and it is the exact line litigation has already tested: a lawsuit filed against OpenAI in the days around this relaunch alleged that ChatGPT advised a user against seeking medical care in a situation that turned out to be serious. Whether that reflects a rare failure mode or a structural risk of grounding a general-purpose chatbot in personal medical context is, as of this rollout, an open question rather than a settled one, and it is the central risk regulators, clinicians, and users will be watching as usage scales from hundreds of millions of queries a week toward whatever comes next.

The wider picture

Health in ChatGPT is one entrant in a fast-forming category of consumer AI tools that want to sit on top of a person's medical history, alongside efforts from device makers, insurers, and health-record platforms doing something similar with their own data. What distinguishes this moment is less the individual feature than the pattern it establishes: sensitive personal data, once portable into a general-purpose AI assistant, is only as protected as that assistant's own policies choose to make it, provider by provider, with no shared regulatory floor underneath. That is also, more broadly, why data portability and not being locked into a single vendor's policies matters for sensitive information generally. A tool like Metir AI, which works across models rather than tying a user's data and workflows to one provider's terms, reflects the same instinct: control over where your information lives is worth as much as any single feature built on top of it.

Looking ahead

Health in ChatGPT is now available to tens of millions of US ChatGPT users, running on a privacy architecture that is more specific than most consumer software offers and a regulatory footing that is thinner than most people assume "medical" data carries. Both facts are true at once, and the six months between the January pilot and the July relaunch suggest OpenAI intends to keep pushing forward rather than pause on the open questions. How that plays out, in complaints, in litigation, in regulatory attention, or in simply becoming normal, will say a great deal about where the boundary between consumer AI and regulated healthcare ends up sitting.

Sources:

Image credits

Header image: Sam Altman, CEO of OpenAI, speaking at an event, by Village Global, CC BY 2.0, via Wikimedia Commons. In-body image: a stethoscope, photographed by Wikimedia user Stethoscopes, CC BY-SA 3.0, via Wikimedia Commons.