ChatGPT can now read patient records in some hospitals
What happened
On 1 September 2026, OpenAI announced that healthcare organisations can connect their electronic health record systems to a product it calls ChatGPT for Healthcare. The connection is to Epic, one of the largest electronic health record systems in the world. Once a hospital sets it up, an authorised clinician can pull a patient's information into ChatGPT and ask questions about it: what has changed since the last visit, what the recent test results show, which medicines were altered, what a specialist recommended. Alongside this, OpenAI launched what it calls a Healthcare Public Data plugin, which links the tool to nine official healthcare data sources, including PubMed, the medical research database, and DailyMed, a US medicines information database. There are two ways it can be used. A clinician can bring record information into ChatGPT, or, in some setups, ChatGPT can be built into the record system itself so the clinician never leaves the patient's chart. OpenAI says it worked with doctors to test the tool across 27 clinical situations, and that those doctors rated 99.1 per cent of the responses as safe. UCSF Health, a large American academic health system, is among the first to pilot it.
Why this matters
Anyone who has sat in a consultation and watched a clinician scroll for two minutes looking for a result will understand the problem this is aimed at. Modern medical records hold an enormous amount of information. Letters, results, scans, medication changes, notes from other teams. Finding the relevant part quickly is genuinely hard, and the time spent looking is time not spent talking to the person in the room. So the appeal is real. If a tool can summarise what has changed since last time, that could hand minutes back to every appointment. The concerns are just as real, and they are not hypothetical. Patient records are among the most sensitive information that exists about a person. Connecting them to a general-purpose AI system raises questions about who can see what, where the data goes, and what happens if the summary is wrong in a way nobody notices. For UK readers, it is important to be clear: this is a US launch, connected to a US record system. It is not something switched on in NHS hospitals. But the NHS is actively adopting AI tools, and the direction of travel is the same.
What the evidence actually says
The safety figure being quoted deserves a closer look, because it is the number most likely to be repeated without context. OpenAI says physician partners rated 99.1 per cent of responses as safe across 27 clinical use cases. That is an evaluation carried out with the company's own partners, not an independent trial, and it measures whether responses were judged safe rather than whether patients ended up better off. Those are different questions. A summary can be safe and still be unhelpful. And 99.1 per cent safe means roughly one in a hundred responses was not rated that way, which in a hospital handling thousands of interactions is not a small number. None of that makes the tool bad. It makes it early. What has not yet been published is independent evidence on accuracy in ordinary use, on how clinicians behave when a summary is wrong, or on effects on patient outcomes. There is also a well-documented pattern worth naming: people tend to trust a confident summary more than they should. The safeguard is a clinician who still opens the record and checks. Whether that happens consistently under time pressure is exactly what nobody yet knows.
Practical advice
You do not need to do anything because of this announcement. Your NHS record is not connected to ChatGPT. But it is a good moment for a few sensible habits. If you want to know how AI is being used in your own care, ask. Surveys of the UK public have consistently found that most people want to be told when AI is involved, and clinicians are generally happy to explain. Keep your own short summary of your health. A single page listing your conditions, your medicines and doses, your allergies, and your recent hospital visits is worth more than any software when you turn up somewhere new. Keep a copy in your bag and a photo on your phone. Check your own record through the NHS App. Many people are surprised by what is in it, and by what is missing or out of date. If something is wrong, tell your practice. And be careful about typing your own medical details into a general chatbot. Once information leaves your control, you cannot get it back.
What to know
OpenAI has connected ChatGPT to Epic electronic health records for healthcare organisations in the United States, letting clinicians ask questions about a patient's record in plain language, alongside a plugin linking it to official medical data sources. The promise is time saved in appointments. The open questions are privacy, accuracy in real conditions, and whether busy clinicians will keep checking. The company's own testing rated 99.1 per cent of responses as safe. That is a reasonable starting point, not independent proof, and it is not the same as showing patients do better. Sources: OpenAI, 'Healthcare organizations can now connect EHR and additional industry data to ChatGPT', 1 September 2026, https://openai.com/index/chatgpt-connects-health-records-and-healthcare-sources/ ; TechCrunch, 'ChatGPT Health adds Epic integration for clinicians to import patient data', 1 September 2026, https://techcrunch.com/2026/09/01/chatgpt-health-adds-epic-integration-for-clinicians-to-import-patient-data/ ; HTN Health Tech News, 'OpenAI introduces EHR and public data integration into ChatGPT for Healthcare', 3 September 2026, https://htn.co.uk/2026/09/03/openai-introduces-ehr-and-public-data-integration-into-chatgpt-for-healthcare/ This article is for general information and does not replace advice from a doctor, pharmacist or other qualified healthcare professional. Do not rely on any AI tool for decisions about your own health or medicines.
Related articles
A knitted fabric that snaps into shape - and works as a switch
Engineers at Harvard have made knitted fabrics that can snap between different shapes and hold them, and that can double as soft electronic switches. The resea…
Scientists have turned bacteria into living transistors
Engineers at MIT have made bacteria behave like the components inside a computer. The work was published in Nature Chemical Biology and reported on 4 September…
An AI now steers fusion plasma faster than any human could
Researchers have built an artificial intelligence system that can steer the superheated gas inside a fusion reactor faster than a human operator could ever reac…