ChatGPT can now read your medical records — should you let it?
On July 23, 2026, OpenAI switched on Health in ChatGPT for US users on every plan — including Free. Connect Apple Health and your patient portals, and ChatGPT stops giving generic health answers and starts answering about you: your labs, your medications, your sleep. This is one of the most useful things ChatGPT has ever shipped, and also the most personal data you will ever hand an AI. Here’s exactly what it does, the privacy fine print in plain English, and a sane way to decide.
Every week 300 million people already ask ChatGPT health questions — now it can answer with your actual records in front of it, and you decide how often it asks permission.
New Health connections now default to your global plugins permission setting, and if you never touched that setting it is “Allow low-risk actions”. In practice: once Health is connected, ChatGPT can use your records when relevant without asking each time; it still asks before anything it treats as sensitive, such as emailing a training plan built on your data to someone. OpenAI says more than 70% of Health users had already chosen this. If you would rather be asked every time — our advice for the first two weeks, and for anyone managing a parent’s records — change it under Settings → Plugins → Health. Existing connections keep whatever you set before.
01 What actually launched
Health is rolling out to logged-in US users 18 and older, on web and iOS, across Free, Go, Plus, and Pro plans. You’ll find it in the left sidebar in ChatGPT (it’s not in Codex). Earlier this year OpenAI tested a separate, dedicated health space with a limited group — and found that over 70% of health conversations happened outside it, mid-way through normal chats about meals, travel, or workouts. So the launch version works everywhere: once you connect your data, any conversation can use it, with your permission.
Two models power it. GPT‑5.5 Instant handles health for Free users and, on OpenAI’s hardest health evaluations, performed comparably to its frontier Thinking models at the time of evaluation. Paid plans get GPT‑5.6 Sol, OpenAI’s strongest health model, built with input from hundreds of physicians.
02 What you can connect
| Source | What ChatGPT gets |
|---|---|
| Apple Health | Sleep, activity, workouts, heart data — plus anything your wearable or fitness app shares into Apple Health. |
| Patient portals | Supported US hospital systems: medications, conditions, lab results, visit notes. |
| One Medical | Primary-care records, synced directly. |
| Function Health | Lab testing results, synced directly. |
After connecting, you review what came in — remove conditions that no longer apply, add family history, correct anything stale. That review step matters: portals often still list a medication you stopped taking two years ago.
03 What it’s actually good for
- Lab results in plain English — “compare this cholesterol panel to my last three” instead of squinting at reference ranges.
- Appointment prep — a one-page summary of what changed since your last visit, plus questions worth asking.
- Decoding visit notes — what the doctor’s note actually says, without the Latin.
- Everyday context — meal plans that respect your allergy, weekend plans that respect your knee. This happens in normal chats, not a special health tab.
You can also type @Health in any message to explicitly pull your health context into a conversation.
04 The privacy fine print, translated
- Not used for training or ads. Connected records, Apple Health data, and the conversations that use them are excluded from model training and ad targeting regardless of your data-control settings.
- Permission by default. ChatGPT asks before using health data in a response. You can approve once, always allow, or change it later in Settings → Plugins → Health.
- Reversible. Disconnect a source in Health → Accounts and its synced data is deleted from OpenAI’s systems within 30 days. What’s already in saved chats stays until you delete those chats.
- Memory stays in your lane. Memories aren’t created directly from your connected records, and no temporary chat ever creates a new memory. One nuance since 27 August 2026: temporary chat is no longer a single setting — a chat you start in personalized mode can still read your existing memories, Custom Instructions and plugins. For a genuinely clean health session, start it non-personalized (that is still the default) and don’t save it to your history afterwards — saving converts it into an ordinary chat. Memory & instructions walks through both choices.
- Extra encryption on health data, on top of standard conversation encryption, plus safeguards before any action that would share health details through another connected app.
This is a real trade, not a trick. The privacy terms here are stronger than ChatGPT’s defaults — but you are still placing your medical history inside a chat app. Start small: connect Apple Health only, live with it for two weeks, and add your medical records only if the answers are earning it. And treat every important answer as preparation for a conversation with your doctor, not a replacement for one — OpenAI itself says ChatGPT can make mistakes.
05 If you help an older parent with their health
This feature may matter most for the person juggling someone else’s health: adult children coordinating a parent’s medications, appointments, and portals. A synced, plain-English picture of “what changed since the last visit” is genuinely valuable there — and so is the caution. Set it up together, review what synced, and set the Health permission to ask every time (it no longer defaults to that). Our AI for Seniors track covers safe setups like this step by step.
Set it up in 10 minutes
- Open Health from the ChatGPT sidebar and choose Get started (US, 18+, web or iPhone).
- Connect Apple Health first — if you use a wearable, check its app is sharing into Apple Health.
- Review what synced: delete stale meds and conditions, add family history that’s missing.
- Set Settings → Plugins → Health to “ask every time” for the first two weeks — since September 14 the default is “Allow low-risk actions”.
- Ask one real question: “Summarize how my labs changed over the past year, in plain English.” Verify one fact against your portal before trusting the rest.
This week’s challenge
Before your next appointment, ask ChatGPT: “Based on my connected records, what changed since my last visit, and what three questions should I ask my doctor?” Bring the list. That’s the feature at its best — you, better prepared, still in charge.