The Promise and Perils of ChatGPT Health

Offers interesting insights, but creates huge data privacy concerns

At a Glance

  • What it is: An experimental feature within ChatGPT (launched January 2026) designed to aggregate your medical records and wellness data.

  • The Promise: It aims to interpret lab results via the b.well network and track daily habits through Apple Health and other apps.

  • The Risk: It is a consumer product, not a doctor. It is not HIPAA-compliant and has a documented history of "hallucinating" medical advice.

  • Privacy Status: It uses a "separate sandbox" to isolate health data from standard chats, but privacy advocates warn that safeguards must be "airtight."



In January 2026, OpenAI released ChatGPT Health, a new platform that promises to make sense of your scattered medical history. For adults 50+ managing complex care, the allure is undeniable: an always-available assistant that can read doctor’s notes and track your sleep.

However, this technology is brand new, largely untested at scale, and should be viewed with a healthy dose of skepticism. While OpenAI describes this as a tool to help you feel "more informed," medical professionals and privacy watchdogs warn that it is strictly experimental and carries significant risks if used incorrectly.

A New Experiment in Your Pocket

ChatGPT Health attempts to merge two disparate worlds: clinical records and daily lifestyle tracking.

1. Clinical Records (The b.well Partnership) Through a partnership with b.well, a health data network, the system allows users to log in to patient portals from trusted U.S. healthcare providers. The goal is to let you upload lab results or visit summaries and ask questions like, "Can you summarize my latest bloodwork?".

  • The Caveat: This relies entirely on the accuracy of the AI's interpretation. While it can summarize data, it is explicitly not intended for diagnosis or treatment.

2. Wellness Tracking The system integrates with Apple Health (for iPhone users only) to track movement, sleep, and activity patterns. It also connects with third-party apps, including MyFitnessPal for nutrition, Peloton for workouts, and Weight Watchers for food guidance.

  • The Caveat: You must manually grant permission for each app. If you disconnect an app, the data stops sharing, but previously discussed information remains in the chat history until you manually delete it.

Privacy: The "Separate Sandbox" Concept

The most critical question for any patient is: Where does my data go? OpenAI has created what they call a "separate sandbox" for this tool.

  • Isolation: Health conversations and files are stored separately from your standard ChatGPT history. If you discuss your blood pressure in the Health section, that data does not flow back into your main, non-health chats.

  • No Training: OpenAI states that your health chats, files, and memories are not used to train their foundation models. Your medical history will not be used to teach the AI how to converse with others.

  • The HIPAA Gap: Crucially, ChatGPT Health is a consumer product, not a medical device. It is not HIPAA-compliant in the way your doctor’s office is, nor is it regulated by health privacy laws.

Privacy advocates, such as the Center for Democracy and Technology, warn that because these tools are not bound by strict medical privacy protections, "inadequate data protections and policies can put sensitive health information in real danger".

Why Skepticism Is Warranted

Despite the glossy launch, this is experimental technology that has been known to fail.

1. The "Hallucination" Problem AI chatbots can confidently state falsehoods. Dr. Robert Wachter of the University of California, San Francisco, noted a case where an AI advised a patient to take ivermectin—an anti-parasitic drug—for testicular cancer. While not immediately fatal, such advice could delay proper treatment. In a more severe documented instance, a user consumed sodium bromide and experienced hallucinations after an AI chatbot gave poor advice on lowering salt intake.

2. False Confidence The system is designed to sound authoritative. Critics argue this "trusted medical adviser" persona can be misleading. While some patients, like cancer survivor Burt Rosen, have successfully used AI to "translate" complex test results into plain English, they represent the best-case scenario of highly tech-literate users.

3. Limited Availability The feature is currently available only to a "small group of early users" via a waitlist, highlighting its beta status. It is not yet available in the UK or Europe, where data protection laws are stricter.

Should You Use It?

My advice is to steer clear of any consumer health app that isn’t HIPAA compliant. Once your private health data is “in the wild” it’s no longer private and you’ve lost control of it. Similar functionality will emerge within the patient portal of your healthcare provider, where it is governed by strict HIPAA privacy rules.

If you’re an early adopter with a burning desire to use this beta technology and you’re willing to put the security of your health data at risk, then proceed with extreme caution.

The Bottom Line: Avoid using ChatGPT Health and wait for HIPAA compliant alternatives. As Dr. Robert Wachter, Chair of the UC San Francisco Dept of Medicine, warns: "The capacity for badness here is pretty high".

Additional Resources

NPR: How patients, and doctors, are using AI to make a diagnosis
https://www.npr.org/2026/01/30/nx-s1-5693219/chatgpt-chatbot-ai-health-medical-advice

Freakanomics Podcast: Can A.I. Save Your Life?
https://freakonomics.com/podcast/can-a-i-save-your-life

Note: this post was prepared with the assistance of Google Notebook LM. The image was generated using Google Nano-Banana.