OpenAI plugs ChatGPT into Epic EHR: nine official healthcare data sources connected as clinical AI reaches hospital core systems

Let's look at how the Epic integration works. OpenAI's official blog describes two complementary workflows. The first is 'EHR context in ChatGPT': authorized patient information from a supported EHR is brought into ChatGPT, letting clinicians review history, identify key changes and prepare for appointments — for example, asking before a visit: 'Which recent lab results should I review before today's appointment?', 'Have there been medication changes or new specialist recommendations?', or 'What follow-ups, referrals, or unresolved issues should I be aware of?' ChatGPT brings together relevant information from the authorized record, summarizes important developments, and points back to supporting chart information. The second is 'ChatGPT in the EHR workflow': in supported deployments, ChatGPT can be embedded directly into the EHR layout, enabling AI-assisted tasks without leaving the patient chart. OpenAI stresses these experiences are designed to 'complement' existing EHR workflows — clinicians still work inside the record system, just with an AI assistant that rapidly synthesizes information. UCSF Health's response confirms the value: as pilot partner, they are exploring how the integration can help clinical teams understand 'what has changed and what matters most' across a complex patient record, reducing time spent synthesizing data by bringing relevant information together more quickly and comprehensively, giving clinicians more time with patients.

Beyond patient records, care teams also rely on external authoritative information — latest research, drug labels, insurance coverage, clinical trials and provider credentials. That is the problem the Healthcare Public Data plugin addresses. It brings dedicated connectors to nine official public healthcare sources, including ClinicalTrials.gov (clinical trial registry), CMS Coverage (Medicare/Medicaid coverage policy), RxNorm (standardized medication terminology), DailyMed (FDA drug labels) and PubMed (biomedical literature). Unlike general web search, these connectors let teams 'work with specific records, fields, identifiers and versions' while focusing tasks on authoritative sources, making it easier to compare and verify precise information — trial eligibility criteria, medication identifiers, coverage policies or drug warnings. OpenAI offers concrete use cases: a research team can use ClinicalTrials.gov to identify actively recruiting trials and compare eligibility criteria; a pharmacy team can use DailyMed to confirm the latest label and warnings for a medication; a population-health team planning a diabetes-prevention program can bring relevant research, active trials and Medicare coverage information together in a source-backed view. These capabilities build on ChatGPT for Healthcare's existing ability to answer clinical questions and synthesize medical research via trusted clinical search across thousands of medical sources.

The most sensitive question in healthcare AI is: how reliable is the model in real clinical contexts? This time OpenAI published fairly robust evaluation data. First, OpenAI works with hundreds of physician advisors across 60 countries, 49 languages and 26 medical specialties to define, measure and improve health-related responses in ChatGPT — to date these physicians have reviewed more than 700,000 model responses reflecting real-world healthcare questions, with feedback improving model behavior and healthcare-specific tools. Second, to understand how ChatGPT performs with connected EHR context, physicians evaluated responses across 27 clinical use cases (including pre-visit review, clinical timelines, medication review and handoff summaries): across 4,363 ratings, 99.1% of responses were rated safe across all use cases. Third, in a separate two-round evaluation, physicians reviewed hundreds of ChatGPT responses to nuanced clinical questions based on large U.S. healthcare datasets: for each of the five connected data sources tested, more than 93% of responses were rated 'good' or better accuracy. OpenAI says these evaluations measure how ChatGPT works with healthcare context — from surfacing relevant information and citing supporting evidence, to preparing work that clinicians and staff can review — and it will keep investing in dedicated healthcare training and evaluation to further improve performance and reliability.

On governance and compliance, the design of this integration is cautious at every level. ChatGPT for Healthcare combines healthcare-specific capabilities with enterprise controls: role-based access control (RBAC), single sign-on (SSO) and audit logs; with an applicable Business Associate Agreement (BAA), customers can use ChatGPT Work, Codex, apps and plugins in the same workspace to support HIPAA-compliant workflows. The read-only design of the Epic integration is itself a risk control — the AI can read and summarize records but has no write-back path, architecturally eliminating the most dangerous scenario of 'AI accidentally modifying a medical record.' Connectors to other enterprise systems (Microsoft SharePoint, Google Drive, Salesforce, Slack plugins) also preserve existing permission structures. On access, OpenAI layers clearly: ChatGPT for Healthcare customers can ask workspace administrators to enable the EHR integration and Healthcare Public Data plugin; ChatGPT Enterprise customers should contact their account team for compliant configuration (Regulated Workspace); and individual clinicians — eligible U.S. ChatGPT for Clinicians users — can only install the Healthcare Public Data plugin, with the EHR integration unavailable for individual accounts. Robert Purinton, Chief AI Officer at AdventHealth, said in a statement that putting tools like ChatGPT Work, Codex and connected business data in team members' hands lets AI reduce routine work and move practical innovations into action faster — the goal is giving caregivers more time for the human connection that enables whole-person care.

Placed in a larger context, this release may well be a watershed for healthcare AI adoption. Over the past two years, healthcare AI has mostly stayed in peripheral scenarios like assisting writing and literature search. But Epic is one of the largest EHR vendors in the U.S. and globally, covering roughly 325 million patient records — plugging ChatGPT into Epic means, for the first time, holding the key to the core of clinical decision-making. For hospitals, the value is immediate: instead of flipping through dozens of pages before a visit, doctors can get a 'what changed, what matters most' summary in under a minute, returning time to patients. For OpenAI, this is a critical step for its enterprise business into a high-moat vertical: healthcare is notoriously hard to enter, and securing Epic means a ticket into thousands of hospitals, paving the way for ChatGPT Work and Codex adoption across medical organizations. The challenges are equally clear: a 99.1% safety rate still means roughly 9 in 1,000 responses may have issues, and healthcare tolerates almost zero error; read-only mode protects records but limits the AI's automation value; and questions of clinician trust in AI summaries, legal liability allocation, and 'who is responsible when AI advice is wrong' remain unanswered. Predictably, more hospitals will follow with pilots, and more debates about misdiagnosis, liability and data privacy will emerge — but one way or another, the door to 'AI in the medical record' has been pushed open.

Looking back at this release, the most notable thing is not 'what AI can do now,' but OpenAI's deliberately restrained landing posture. On capability, it could have let AI automatically organize records, generate documentation or even offer treatment suggestions — yet it deliberately narrowed the first step to 'read-only + summarize + assist.' On scope, it did not open the floodgates to every hospital, instead building a slow, steady entry path with pilot partners (UCSF Health), tiered permissions (organization/enterprise/individual) and compliance-first requirements (BAA, HIPAA, Regulated Workspace). Behind this restraint is a clear-eyed view of healthcare's nature: medical records are hospitals' most valuable asset and liability; clinicians are instinctively wary of 'AI deciding for me'; and regulators tolerate almost zero error in medical AI — the first ticket into this market is not technical flash, but trust. OpenAI's strongest trust assets are precisely the 700,000 physician-reviewed responses, the 99.1% safety rate, and the architectural promise of 'read-only, no write-back.' From ChatGPT for Healthcare's January launch, the Clinicians version in April, to today's Epic integration, OpenAI's healthcare roadmap is clear: refine the model with physician reviews first, then break through the key nodes of clinical workflows one by one. For industry observers, the metrics worth tracking next are: pilot hospitals' usage and physician satisfaction, the real incidence of false positives and misses, and whether EHR vendors beyond Epic follow suit — if this model of 'read-only access + authoritative data sources + physician review' proves viable, it may well become the standard posture for healthcare AI entering clinical practice.