ChatGPT’s Health Push Repeats a Bet That Killed Two Rivals

OpenAI switched on Health in ChatGPT for every adult US user this week, letting the chatbot read Apple Health data and hospital medical records on request. The company says more than 300 million people already ask it health questions every week.

Google and Microsoft tried something similar. Both quit. OpenAI is now walking the same ground carrying an audience its predecessors never had, and a subpoena from 42 state attorneys general that they never faced either.

What ChatGPT Can See Once You Connect Your Data

With permission, ChatGPT can now pull in medications, lab results, recent visits, sleep and activity data whenever a user asks a health question, not just inside a separate health tab. It can compare a new lab result against an old one, summarize what changed since the last appointment, or connect a bad night’s sleep to a missed workout streak.

OpenAI frames the goal narrowly: cut down on retyping the same medical history into every conversation, and support, not replace, a real doctor. The company laid out the mechanics in its own announcement of the rollout, which covers Free, Go, Plus and Pro plans on web and iOS for logged-in users 18 and older. It is not available in Codex, and OpenAI has not given a timeline for Android or countries outside the US.

Two model tiers now sit behind those conversations, split by plan.

The Health tab in the sidebar stays the control center: connect sources there, browse synced records, track trends over time and revisit old health conversations.

Why 300 Million Weekly Questions Changed OpenAI’s Math

Earlier this year OpenAI tested a dedicated health space with a limited group of users. It found something that reshaped the whole rollout: more than 70% of health-related conversations were happening outside that space entirely, scattered through ordinary chats.

People were already asking. OpenAI’s data just was not there to answer with. The company says health information about any one person is normally scattered across patient portals, paper records, fitness apps and wearables, none of which talk to each other.

That is the gap Health in ChatGPT is built to close, and it explains why OpenAI chose to open the feature to everyone rather than keep testing a walled-off version. A habit already existed. OpenAI just had to decide whether to feed it real data.

Two Data Graveyards Named Google and Microsoft

OpenAI is following a well-worn path. Two funded, credible predecessors tried to become the place people store medical history, and both gave up.

1. 2007: Microsoft launches HealthVault, a free online store for medical records, betting patients and hospitals will fill it.
2. 2008: Google launches Google Health, letting users import prescriptions and history from partnered pharmacies and hospitals.
3. 2011: Google announces it will close Google Health, citing failure to reach broad adoption despite hospital and insurer partnerships.
4. 2018: Apple launches Health Records, eventually signing more than 140 hospital and health system partners.
5. 2019: Microsoft shuts down HealthVault after twelve years, pointing to low usage and a pivot toward enterprise software.
6. 2026: OpenAI opens Health in ChatGPT to every US adult, betting its weekly health questioners succeed where dedicated portals did not.

Both companies died from indifference. Adoption never reached critical mass, in hospitals or at home, and neither ever had to fight off a privacy backlash because barely anyone showed up long enough to trigger one.

Adoption is solved for ChatGPT already. What is left open is whether the medicine underneath actually holds up, and OpenAI has spent months trying to answer that with physicians rather than engineers alone.

Does ChatGPT Know Medicine Better Than Doctors

OpenAI says GPT-5.6 Sol scored 60.5 against 43.7 for physicians’ own written answers on HealthBench Professional, a 525-task exam built with practicing doctors. The test measures written medical reasoning, not hands-on diagnosis, and OpenAI still tells users to confirm anything important with a licensed clinician.

OpenAI shipped the GPT-5.6 family on July 9, in two versions, Sol built for depth and Luna built for speed, two weeks before Health’s nationwide rollout. The company says the entire GPT-5.6 lineup, Sol included, outperformed GPT-5.5 models on HealthBench Professional.

The benchmark itself leans on a large physician network: 525 physician-authored tasks spanning consults, clinical writing and medical research, built by 260 doctors across 60 countries and 26 specialties who reviewed more than 700,000 model responses while refining the scoring rubrics. OpenAI publishes the underlying scores on its own deployment safety documentation for GPT-5.6, which grades models on accuracy, safety, communication, context awareness, completeness and knowing when to escalate to a human.

The Records Just Lost Their HIPAA Shield

Here is the part Google Health and HealthVault never had to reckon with: once a medical record leaves a hospital, clinic or insurer and lands inside a consumer chatbot, HIPAA’s federal privacy rules stop applying. The user is left governed by OpenAI’s own terms of service instead.

Sharing your electronic medical records with ChatGPT Health would remove the HIPAA protection from those records, which is dangerous.

Sara Geoghegan, senior counsel at EPIC, the Electronic Privacy Information Center, made that point days after the nationwide launch. The group warned that the shift strips medical records of federal protection once they leave a hospital or insurer’s systems.

EPIC also argued that OpenAI is bound only by its own disclosures, not by a health privacy statute, so the company could revise its terms at any time. The group pointed to 23andMe as the cautionary case: a firm can hold firm privacy commitments right up until a bankruptcy or acquisition changes who actually owns the data.

OpenAI’s counter is that connected medical records, Apple Health data and any conversation using them are excluded from foundation model training and ad targeting, regardless of a user’s general model-training setting. That exclusion carries more weight now that OpenAI has also begun showing ads to free-tier ChatGPT users elsewhere in the app, a separate revenue push that makes the health data carve-out a real promise rather than a formality.

Forty-Two State Attorneys General Are Circling

Meta made a similar privacy pitch this year, testing a private AI chat mode it says even it cannot access. Trust claims like these are becoming standard across the industry, and regulators are no longer taking them at face value.

A coalition of 42 state attorneys general, led by New York’s Letitia James, subpoenaed OpenAI on June 12, 2026. The document demands cover advertising claims, user engagement and retention design, consumer health data, how the company treats minors and older users, model specifications and complaints about AI sycophancy.

It is described as the broadest coordinated state legal action against an AI company in US history. The subpoena landed four days after OpenAI filed a confidential S-1 registration statement, reportedly targeting a valuation of up to $1 trillion, meaning the same regulators now hold documents on health data handling just as OpenAI courts public investors.

Connecting and Disconnecting Your Data

Setting it up follows a guided flow rather than a single toggle.

• Open Health from the ChatGPT sidebar or the More menu and select “Get started.”
• Connect Apple Health, a supported US hospital system, One Medical or Function Health, then allow a few minutes for the initial sync.
• Confirm the relevant apps are actually sharing data into Apple Health, since ChatGPT can only read what Apple Health itself receives; some proprietary scores never transfer.
• Review synced conditions and medications, remove anything outdated, and add family health history ChatGPT would not otherwise see.
• Add @Health to any message to pull that context into a specific conversation on demand.

By default, ChatGPT asks permission before using connected records to personalize an answer; users can approve requests one at a time or flip on “always allow” under Settings, Plugins and Health, which turns the prompts off. Disconnecting a source under Health, Accounts triggers deletion within 30 days, though anything already saved inside past conversations sticks around until those conversations are deleted separately. Memories are not created directly from connected records, and Temporary Chat or a disabled memory setting keeps a session out of that system entirely.

Google’s and Microsoft’s health records died from indifference. ChatGPT’s version launches with hundreds of millions of built-in users already asking, and a 42-state subpoena already on file.

Frequently Asked Questions

Is Health in ChatGPT Covered by HIPAA?

No. Once medical records leave a hospital, clinic or insurer and enter a consumer app like ChatGPT, HIPAA’s federal privacy rules no longer apply. Users are instead governed by OpenAI’s own terms of service and privacy policy, the core objection EPIC raised once the nationwide rollout began.

What Is HealthBench Professional?

It is OpenAI’s internal test for medical reasoning, built from 525 tasks written by physicians across three areas: patient consultations, clinical writing and documentation, and medical research. The underlying physician network spans 260 doctors in 60 countries and 26 specialties, who reviewed more than 700,000 model responses while building the scoring rubrics.

Will Health in ChatGPT Come to Android or Countries Outside the US?

OpenAI has not announced a timeline for either. The feature currently works only on web and iOS for US-based adults, echoing how Apple Health Records itself has stayed tied to the iPhone since it launched in 2018.

Does Disconnecting a Health Source Delete My Data Right Away?

Not instantly. OpenAI says data from a disconnected source is removed from its systems within 30 days, but information already saved inside past conversations stays until a user deletes those conversations separately.

Disclaimer: This article is informational only, not medical or legal advice; consult a healthcare provider before connecting personal medical records to any AI service, and note that figures here reflect OpenAI’s public statements as of publication.