ChatGPT is now plugged into the medical records of 325 million patients

Key points

• OpenAI has integrated ChatGPT Health with Epic, an electronic health record (EHR) system that holds data for more than 325 million patients.
• Clinicians get read-only access: the AI can read and summarise records but cannot write anything back to them.
• OpenAI surveyed 4,300 physicians across 27 clinical tasks and found 99.1% of AI responses were rated safe.
• A new Healthcare Public Data plugin pulls information from sources including ClinicalTrials.gov and PubMed to help staff check drug details and trial eligibility.
• Two lawsuits against OpenAI allege that ChatGPT gave harmful medical advice to individual users.

For most people, a doctor's appointment is a blur of paper notes, old test results and half-remembered medication names. OpenAI wants to fix that, at least from the clinician's side.

The company announced this week that ChatGPT Health now connects directly to Epic, an electronic health record system, the software hospitals use to store everything from appointment notes to lab results. Epic holds records for more than 325 million patients, which makes it the largest such system in the United States.

What can doctors actually do with this?

Clinicians can pull up a patient's history, lab results, medications, and specialist notes, then ask ChatGPT to summarise them in plain language. In some hospitals, they can do all of this without ever leaving the patient's chart screen.

The AI builds pre-visit summaries, flags changes in a patient's condition over time, and helps prepare handoff notes, the brief a nurse or doctor writes when passing a patient's care to the next shift. OpenAI is clear that the integration is read-only: ChatGPT cannot edit, add to, or delete any record. It reads; it does not write.

OpenAI also launched a Healthcare Public Data plugin, a tool that fetches information from sources including ClinicalTrials.gov (a database of medical research studies), RxNorm (a standard list of drug names), DailyMed (official drug label information), and PubMed (a library of medical research papers). Clinicians can use it to check whether a patient might qualify for a clinical trial or to verify medication details quickly.

Should patients be worried?

OpenAI says it surveyed more than 4,300 physicians across 27 types of clinical task and found that 99.1% of the AI's responses were rated safe. That figure sounds reassuring, but the company itself acknowledges that AI is not suitable for diagnosis or treatment, and even a small error rate in medicine can cause real harm.

The timing is uncomfortable. Just days before this announcement, a Florida pastor filed a lawsuit claiming ChatGPT gave him a near-fatal recommendation. In May, the family of another user sued OpenAI, alleging the chatbot gave dangerous advice about medication dosage. OpenAI has not commented publicly on the specifics of either case.

Last month the company opened ChatGPT Health to all US consumers, and it says people already send 300 million health-related questions to ChatGPT every week.

What happens next?

Organisations that sign a Business Associate Agreement, a legal contract required under US health-privacy law, can now also use ChatGPT for work tasks, coding tools, and third-party app connectors inside their systems.

If your hospital uses Epic, your doctor may already have access to these tools. The safeguard to know: the AI reads your records to help your clinician prepare, but it cannot change what is written there.