What happened
On July 23, 2026, OpenAI published an announcement titled "Launching Health in ChatGPT." OpenAI's own summary says the feature "now lets eligible U.S. users securely connect medical records and Apple Health to get more personalized insights and better understand their health." The full OpenAI page could not be retrieved for this briefing, so anything attributed to OpenAI here comes from that summary or from press-briefing reporting.
TechCrunch reported the same day that the feature is reaching U.S.-based users aged 18 and over across the free, Go, Plus, and Pro plans, on the web and iOS. The Verge reported the same eligibility, and added that health questions now work in the main chat rather than a separate Health tab, and that users can review and change which data types they link, such as lab results, medications, and post-visit notes. Between the two reports, connectable services include Apple Health, MyFitnessPal, and Function, alongside records from provider systems including Epic and Oracle Health.
Several claims belong to OpenAI rather than to independent testing. The company told reporters that conversations are encrypted in transit and at rest with additional protection for Health data, that it does not use this data to train its models, and that a smaller model scores better than its predecessor on a health benchmark. The Verge reported that OpenAI's vice president of health product said its models reason "at levels that are better than clinician level," and that OpenAI's health lead said he would "temper" that claim.
Why it matters
Connecting records changes the shape of the risk. A general assistant guessing from a vague description is often wrong in a harmless way. A chatbot that has read your lab results and medication list sounds specific and authoritative, and a confident wrong answer about a symptom or a drug interaction is a different kind of mistake.
It helps to separate what is established from what is asserted. Established: the feature exists, it is rolling out in the U.S., and it can connect records and fitness data. Asserted by the vendor and unverified here: the security posture, the no-training promise, and the performance numbers. Scores from AI evals measure a test set, not safety in your situation, and OpenAI's own health lead walked back the clinician comparison. Both outlets also noted a lawsuit filed in Florida the day before, alleging ChatGPT gave dangerous medical recommendations that led to delayed care. Those are allegations that have not been tested in court. Nothing in the sources suggests this is cleared for diagnosis; OpenAI describes it as supporting, not replacing, professional care.
What to do next
- Treat it as an information tool, not care. For anything urgent, such as chest pain or breathing trouble, contact emergency services or a clinician.
- Do not start, stop, or change a medication or treatment because of an AI answer. Confirm with the clinician who knows your history.
- Connect the minimum. You can choose which data types to link, so start narrow.
- Read the privacy and retention terms yourself before linking records. The encryption and training statements here are OpenAI's own, not confirmed by this briefing.
- Check any answer against a second source, and bring it to an appointment as a question, not as a conclusion.
This briefing summarizes OpenAI's July 23, 2026 announcement summary and same-day reports from TechCrunch and The Verge. OpenAI's full announcement page could not be retrieved here, capability, privacy, and security statements are the company's own and were not independently verified, the lawsuit described is an unproven allegation, and nothing here is medical advice.