Is ChatGPT Health private? Records, memory, training, and HIPAA

The Morse teamPublished November 5, 20267 min read

As of November 2026, ChatGPT Health is more private than an ordinary ChatGPT chat, but it isn't private from OpenAI, and HIPAA doesn't cover it. OpenAI says connected medical records and Apple Health data, and the conversations that use them, aren't used to train its foundation models or target ads, and that they get "additional encryption protections" (OpenAI). OpenAI doesn't call Health end-to-end encrypted, and its systems read your records to answer, so the copy on its servers is one OpenAI can read. Since July 23, 2026, Health is no longer the walled-off space OpenAI announced in January.

This isn't medical or legal advice. We read OpenAI's Health help page, its January 7 and July 23, 2026 launch posts, its release notes, its Health Privacy Notice, its US privacy policy, and its parental controls and app permission pages on November 5, 2026, along with the US Department of Health and Human Services (HHS) page on health data in personal apps. OpenAI deserves credit for one promise. Its training exclusion for connected records holds "regardless of the model-training setting you choose," which no typed ChatGPT chat gets.

ChatGPT Health privacy at a glance

OpenAI renamed the feature "Health in ChatGPT" in July 2026. Each row comes from OpenAI's own pages, read November 5, 2026.

Is ChatGPT Health HIPAA compliant?

No. OpenAI's help page says Health in ChatGPT "is not intended for clinical or covered-entity use and does not offer a Business Associate Agreement," and sends hospitals and insurers to ChatGPT for Healthcare instead (OpenAI). Records you export from a patient portal into a consumer app you chose yourself sit outside HIPAA, as we explain in does HIPAA apply to ChatGPT.

Does ChatGPT Health train on your medical records?

Not on the records you connect, or on the chats that draw on them, though a health question typed without connected data falls under your usual training switch, as our health questions guide lays out from OpenAI's help page.

The Health Privacy Notice, updated June 29, 2026, words the promise differently from the help page. The notice says that "by default" OpenAI doesn't use Health data to improve its foundation models. The same paragraph adds that "a limited number of authorized OpenAI personnel and trusted service providers might access data received via Health Features to improve model safety, unless you have opted out" (OpenAI). The help page has no sentence on people reading Health data.

Is ChatGPT Health still a separate space?

No. It was separate when it launched, and OpenAI changed that in July. OpenAI's launch posts, release notes, and policy pages give the dates.

Many guides still describe the January design, with a separate space and its own memory. OpenAI's July 23 post said that "by default, ChatGPT asks for permission" before using your records. By September 14, new connections had stopped asking by default. To get the prompt back, choose Always ask under Settings, then Plugins, then Health (OpenAI).

The old separation survives in one place. OpenAI's help page says chats from the earlier Health experience sit in a Health Project that "retains the segmented controls," so its memories "cannot be used to personalize your chats elsewhere in ChatGPT" (OpenAI). New Health chats don't get that by default.

Does ChatGPT Health remember what you tell it?

Yes, when memory is on. OpenAI's help page says "conversations that use Health can create memories to personalize future chats," and gives the example of a sleep goal that ChatGPT keeps in mind later (OpenAI). ChatGPT doesn't create memories directly from the records you sync. What you say about those records in a chat can become a memory.

OpenAI offers three ways to keep a health chat out of memory. A Temporary Chat creates no memories, though OpenAI's privacy policy says Temporary Chats are deleted "within 30 days" rather than at once (OpenAI). Memory can be turned off under Settings, then Personalization. A separate project with project-only memory keeps Health memories from mixing with the rest. OpenAI's memory page warns that "deleting a chat alone does not necessarily delete a separate saved memory created from that chat" (OpenAI), and our guide on how to turn off ChatGPT memory covers the steps.

Is ChatGPT Health encrypted?

Yes, in transit and at rest. OpenAI's help page says "all chats in ChatGPT are encrypted by default at rest and in transit, and information connected in Health receives additional encryption protections" (OpenAI). The January post called these "purpose-built encryption and isolation" (OpenAI).

Neither page explains what the additional protections are or who holds the keys. OpenAI doesn't call Health end-to-end encrypted, and its systems read your records to answer, so the copy on its servers is one OpenAI can read. The difference is explained in how end-to-end encrypted AI chat works.

What happens to your records when you disconnect or delete?

Disconnecting a source under Health, then Accounts, deletes what it synced within 30 days, but any chat that already quoted those records stays until you delete it (OpenAI).

Deleting the chats starts a second clock. OpenAI's privacy policy, updated September 10, 2026, says deleted data is removed "within 30 days," and that if OpenAI receives "a lawful subpoena," it "may retain it for the duration of the relevant legal or regulatory obligation" (OpenAI). The Health Privacy Notice lists the same legal grounds for disclosing Health data, and says OpenAI does "not sell Personal Data" collected through Health. How courts have reached ChatGPT chats is in can AI chats be subpoenaed.

How Morse keeps your health questions private

Each safeguard in ChatGPT Health governs the copy that OpenAI stores under your account, and OpenAI rewrote the separation promise twice between January and September 2026. On Morse, the history of a chat lives on your own device, so there's no copy on Morse's servers for a policy change to reach.

• E2EE, for a test result, a diagnosis, or a prescription you'd show no one. Before your message leaves the device, your browser verifies the enclave's hardware attestation, then seals the text for that one enclave. What crosses Morse's servers is ciphertext they can't read, and the answer comes back encrypted for your browser alone. Third-party providers run the enclaves, not Morse, and the protection depends on the enclave hardware and attestation working as designed. Tap the lock on a reply to see every check it passed, or download the full report.
• Private, the default, for routine questions about sleep, diet, or a side effect. Your message goes to open-weight models such as GLM, DeepSeek, and Kimi, or to xAI's Grok, with your identity removed. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing, and Morse rejects any reply that lacks the zero-retention confirmation where a provider sends one.
• Anonymous, when you want GPT, Claude, or Gemini to answer without learning who asked. The model's maker reads the question under Morse's account, minus your email, account, and IP address, and may keep it for a limited time to check for abuse, under its own policy. Leave out names, birth dates, and record numbers here.

Morse makes no HIPAA claim. A chat's history sits in your browser, encrypted, and opening it needs a key held on your device plus a second key our server hands over when you sign in. Morse doesn't train AI on what you write, sell your data, or show ads. With no chats on our servers, we usually have very little to give authorities (privacy policy). The flip side is that a cleared browser, a lost device, or a deleted account takes the chats with it beyond Morse's reach, so copy out the notes you want for your next appointment.

So is ChatGPT Health private enough for you?

• Training: connected records are excluded in ChatGPT Health, but typed questions follow your setting. Morse doesn't train AI on what you write.
• Records used without asking: set Settings, then Plugins, then Health to Always ask. On Morse, a lab report you attach is sent with that one message and never stored by Morse.
• Health details showing up in other chats: use a project with project-only memory, or a Temporary Chat. On Morse, your history stays in your browser.
• Anything you'd keep from OpenAI or a reviewer: start the chat in E2EE on Morse.

Try a chat from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.