TL;DR
- OpenAI’s Epic integration brings authorised patient context into ChatGPT for Healthcare, with ChatGPT available inside supported EHR layouts too.
- A separate plugin connects to nine official healthcare sources for research, medication, coverage and related reference work.
- The workspace includes organisational controls; start with a defined use case, appropriate permissions and clinician review against the source information.
- OpenAI reports separate EHR safety and public-data accuracy evaluations. The Epic integration is not available to individual accounts.
In plain English
An electronic health record, or EHR, holds patient information such as notes, lab results and prescriptions. Epic supplies these systems to healthcare organisations.
Structured access lets software work with particular records, fields and identifiers rather than relying only on a general text search.
Role-based access limits what someone can use according to their role. Audit logs record activity so an organisation can review what happened.
A Business Associate Agreement, or BAA, sets out responsibilities for handling protected health information under US HIPAA rules. It works alongside technical controls and the organisation’s procedures.
OpenAI has introduced an Epic integration for ChatGPT for Healthcare, alongside a plugin connecting it to nine official healthcare data sources. Together, they bring patient information and public medical reference material into the workspace where healthcare teams use ChatGPT.
For a clinician, the immediate uses are familiar: preparing for an appointment, reviewing a patient’s history, checking for medication changes and identifying unresolved follow-ups. The integration is designed to gather relevant information from an authorised patient record, summarise it and point back to the supporting chart information.
That connection changes what an assistant can work from. A general clinical question can be answered from medical reference material. A question about what changed for a particular patient needs their record: appointment notes, lab results, prescriptions and specialist documentation. Bringing that context into the workflow gives the clinician a way to ask across those sources while retaining a route back to the chart.
OpenAI describes two ways to use it. Authorised EHR context can be brought into ChatGPT, and supported deployments can place ChatGPT directly inside the EHR layout. The second approach lets staff use the assistant without leaving the patient chart. Which experience is available depends on the deployment.
The Healthcare Public Data plugin covers a different part of the work. It connects to nine official sources, including PubMed, DailyMed, ClinicalTrials.gov and CMS Coverage. OpenAI describes structured access to records, identifiers, fields and versions, so a task can stay focused on specific reference information.
A research team could compare trial eligibility criteria; a pharmacy team could check medication labelling and warnings; a planning team could bring research and coverage information together. These are examples in the announcement, illustrating how the plugin is intended to help teams work across sources they would otherwise consult separately.
The broader direction is a workspace that supports clinical, research and administrative work with access to the appropriate information. OpenAI also describes ChatGPT Work for preparing reports, analyses and plans, Codex for software work, and plugins for business systems. The practical value depends on fitting those capabilities into the organisation’s existing work and permissions.
OpenAI names role-based access, single sign-on and audit logs among the controls. With an applicable Business Associate Agreement, it says customers can use the workspace to support HIPAA-compliant workflows. Having several capabilities in the same workspace does not mean every one can access patient data; the organisation’s configuration and access rules remain important.
The announcement includes physician evaluations of the connected workflows. For EHR work, OpenAI reports 4,363 ratings across 27 clinical use cases and says 99.1% of responses were rated safe. In a separate evaluation of public-data work, more than 93% of responses received a “good” or better accuracy rating for each of five sources tested.
Those results describe different samples and measures, so they should stay attached to the workflows they assessed. The announcement does not specify the distinct response count behind the EHR ratings or provide its per-use-case breakdown. Teams considering a deployment should use the reported results as background for checking the tasks they intend to run, including whether summaries preserve important chart details and make the supporting evidence easy to review.
Access is organisational for the Epic integration. ChatGPT for Healthcare customers can ask their administrator to enable it; Enterprise customers need to confirm eligibility and configuration with OpenAI. Eligible US ChatGPT for Clinicians users can install the public-data plugin, but individual accounts do not get the EHR connection.
For a healthcare team, a sensible starting point is one defined workflow, such as pre-visit review, with the appropriate permissions and a clear way for staff to check the result against the record. The significance of the release is that the assistant can work closer to the information and systems clinicians already use. The next step is making that connection useful in the actual clinical workflow.