ChatGPT to connect with nine official medical databases and Epic EHR

The target is Epic, the EHR software used by many large U.S. hospitals. Once a hospital links its Epic environment to a ChatGPT for Healthcare account, physicians can skip the manual hunt through appointment notes, lab results, medication records, and specialist opinions, and instead ask ChatGPT directly: "What's changed since the last visit?" or "Are there any recent test results I should review before today's appointment?"

OpenAI says it's supporting this in two ways. One lets clinicians pull up approved patient information inside the ChatGPT interface to review medical history. The other embeds ChatGPT directly into Epic's own screen layout, so clinicians get AI assistance without ever leaving the patient chart. Both approaches are designed to layer ChatGPT on top of existing EHR workflows rather than replace them.

UCSF Health served as a pilot partner for this integration. Suresh Gunasekaran, CEO of UCSF Health, said the technology "has the potential to reduce the time it takes to synthesize data."

Nine official medical databases, one plugin

Patient records alone aren't enough. Clinicians also need to check the latest research papers, drug approval information, insurance coverage criteria, and clinical trial enrollment status — information that's typically scattered across separate websites. The healthcare public data plugin OpenAI released alongside the integration bundles nine official sources into one, including ClinicalTrials.gov (trial registrations), CMS Coverage (Medicare coverage criteria), RxNorm (standardized drug codes), DailyMed (drug labeling information), and PubMed (medical literature search).

OpenAI's examples show a wide range of uses. Research teams can search ClinicalTrials.gov for enrolling trials and compare eligibility criteria. Pharmacy teams can pull up the latest label and warning information for a specific drug via DailyMed. A public health team designing a diabetes prevention program could pull relevant research, ongoing trials, and Medicare coverage information into one view, complete with supporting evidence, OpenAI said.

Physicians reviewed over 700,000 cases to score safety

For AI handling medical information, accuracy is what builds trust.

The announcement included results from two separate evaluation tracks. One assessed 27 clinical scenarios in the EHR-connected context — pre-visit review, clinical timelines, medication review, handoff summaries, and more. The other measured accuracy on complex questions drawn from large U.S. medical datasets.

Evaluation	Scope	Result
Clinical scenario safety	27 use cases, 4,363 evaluations	99.1% rated safe
Accuracy by data source	5 connected data sources	93%+ rated "good" or better

Physicians judged the results based on how well ChatGPT surfaced relevant information, how accurately it cited sources, and whether the output was something a clinician could reasonably review and act on.

How to get access

Access depends on the type of account.

User type	How to access	Scope of access
ChatGPT for Healthcare customers	Request EHR integration and plugin activation from workspace admin	Both EHR integration and plugin
ChatGPT Enterprise customers	Contact OpenAI account team to confirm regulated workspace setup	Access after eligibility confirmation
Individual clinicians (U.S.)	Install plugin if eligible for ChatGPT for Clinicians	Healthcare public data plugin only; no EHR integration

In a hospital workspace, an administrator has to turn on both features first, after which individual clinicians can start pulling in patient context or searching the nine sources within ChatGPT. Individual clinician accounts only get access to the plugin — EHR integration, which involves patient personal data, is only available under organizational contracts.

OpenAI says the same workspace also supports using ChatGPT Work to turn clinical and administrative information into reports and analysis, and using Codex to build clinical support software. Plugins connecting to Microsoft SharePoint, Google Drive, Salesforce, and Slack can also be added to the same workspace while preserving existing permission structures.

Editor's take

What OpenAI chose to emphasize here isn't a feature list — it's the scale of validation. For ChatGPT to go from a general productivity tool to something embedded in patient charts, it first has to close the gap between an answer that sounds plausible and one that's actually clinically safe. In a domain where a single error can directly affect a treatment decision, OpenAI led with these validation numbers rather than the feature announcement itself, clearly trying to earn trust before anything else.

ChatGPT's memory feature was built to remember an individual user's task context. This Epic integration is a different animal entirely — an enterprise feature tangled up with organizational contracts, HIPAA business associate agreements, and role-based access controls. It's a sign that under the same ChatGPT brand, consumer convenience features and regulated-industry features are now evolving at increasingly different speeds.

For hospitals and health IT teams in Korea, it's worth separating two things. One is the healthcare public data plugin, which works independently of any specific hospital's EHR. The other is the EHR integration, which is tied to one specific vendor: Epic. Since most large hospitals in Korea run domestic EHR systems rather than Epic, the number of hospitals that could adopt this integration as-is is likely limited. That said, the plugin's approach — pulling clinical trial, drug, and coverage information into one place — is a design worth studying regardless of which EHR a hospital uses.

The thing to watch over the coming weeks is whether OpenAI announces integration plans with EHR vendors beyond Epic. Supporting Epic alone, even though it covers a large share of U.S. hospitals, won't be enough to reach the global EHR ecosystem — and how fast that expansion happens will determine how widely ChatGPT for Healthcare actually gets adopted.