Ninety-nine percent. That’s the share of physician-rated responses that OpenAI’s ChatGPT for Healthcare scored as safe across 4,363 clinical evaluations tied to connected EHR context. For a product category still earning the trust of health systems, that number matters. And it’s the backdrop to what OpenAI announced on September 1, 2026: a direct integration with Epic and a new Healthcare Public Data plugin that together pull authorized patient records and official clinical datasets into ChatGPT’s governed workspace.
This isn’t AI bolted onto a clinical workflow. It’s an attempt to meet clinicians where the information already lives, and reduce the time spent toggling between systems to piece together a coherent clinical picture.
What the Epic integration actually does
The core problem it addresses is familiar to anyone who works in a busy clinical setting. Patient information is scattered: appointment notes in one place, lab results in another, specialist letters somewhere else. Before a clinic visit, pulling it all together is time-consuming work that often falls on already-stretched clinical staff.
The Epic integration lets clinicians query authorized patient records directly inside ChatGPT. They can ask what has changed since a patient’s last visit, which lab results need attention before today’s appointment, or whether there are unresolved referrals or follow-ups outstanding. ChatGPT surfaces relevant information from the chart, summarizes key developments, and links back to source documentation.
The integration runs in two modes. Clinicians can bring Epic patient context into ChatGPT for review alongside other information sources, or, in supported deployments, ChatGPT can be embedded directly into the EHR layout so the workflow never leaves the chart. UCSF Health is a pilot partner. Its president and CEO, Suresh Gunasekaran, described the goal as helping clinical teams understand what has changed and what matters most across complex records, faster and more completely than current workflows allow.
Nine official data sources, one plugin
Patient records are only part of what clinical teams need. They also rely on current research, drug labeling, coverage policy, and trial eligibility criteria. The Healthcare Public Data plugin addresses this by connecting ChatGPT to nine official sources through dedicated, structured connectors. Users can work with specific records, fields, and version-controlled information without hunting through each source separately.
The nine connected sources include:
- PubMed
- ClinicalTrials.gov
- DailyMed
- RxNorm
- CMS Coverage
- And four additional official public health datasets
A pharmacy team could use DailyMed to confirm current labeling and warnings for a specific medication. A research coordinator could cross-reference trial eligibility criteria on ClinicalTrials.gov. A population health team building a diabetes prevention program could pull together relevant published evidence, active trials, and Medicare coverage data in a single source-backed view. That kind of synthesis, done manually, can take hours. Done here, it’s a query.
How the safety and accuracy evaluations were run
OpenAI says it works with hundreds of physicians across 60 countries, 49 languages, and 26 specialties to evaluate model responses. More than 700,000 responses have been reviewed to date. For the EHR-specific evaluation, physicians rated responses across 27 clinical use cases including pre-visit summaries, medication review, and handoff documentation. The 99.1% safety rating came from those 4,363 ratings.
A separate accuracy evaluation tested responses against nuanced clinical questions drawn from large U.S. healthcare datasets. Across five connected data sources, more than 93% of responses were rated “good” or better for accuracy. These are physician-led evaluations, not automated benchmarks, which gives them more clinical weight than standard model leaderboards.
Beyond clinical use: operations, research, and IT
The same workspace also connects to enterprise tools. Plugins for Microsoft SharePoint, Google Drive, Salesforce, and Slack expand the business context available to administrative and operational teams, while preserving existing permission structures. OpenAI’s Codex is available for technical teams building or improving clinical software. AdventHealth’s Chief AI Officer, Robert Purinton, framed the value simply: reduce routine work so care teams have more time for patients.
Availability and compliance
ChatGPT for Healthcare includes role-based access controls, single sign-on, and audit logs. With an applicable Business Associate Agreement in place, customers can use the workspace for HIPAA-compliant workflows. Existing ChatGPT for Healthcare customers can ask their workspace administrator to enable the Epic integration and Healthcare Public Data plugin. ChatGPT Enterprise customers should contact their OpenAI account team to discuss access.