Skip to content
NewsProduct launch

OpenAI Connects Epic EHR and Nine Healthcare Databases to ChatGPT for Clinicians

· by Pondero Newsdesk

The short version

ChatGPT for Healthcare can now pull read-only patient context from Epic and query nine official public health databases, putting frontier AI directly inside clinical record workflows.

OpenAI Connects Epic EHR and Nine Healthcare Databases to ChatGPT for Clinicians

Clinicians at seven named U.S. health systems can now open ChatGPT and pull authorized patient context directly from Epic, without leaving the ChatGPT workspace. OpenAI made the Epic EHR integration and a nine-source Healthcare Public Data plugin available on September 1, 2026, per the OpenAI announcement. The move marks the first time a frontier AI chat product has connected directly to clinical-record systems at the patient-data level under an enterprise governance framework.

What

The Epic integration gives ChatGPT for Healthcare the ability to surface appointment notes, laboratory results, medications, and specialist documentation from a patient's authorized chart, per the OpenAI announcement. Access is read-only; the model cannot write anything back to the EHR. In supported deployments, ChatGPT can also be embedded directly inside the EHR layout, so clinicians do not need to switch applications to run a query.

The Healthcare Public Data plugin connects ChatGPT to nine official sources including ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed, and PubMed, per the same announcement. Teams can compare trial eligibility criteria, verify medication identifiers, and cross-reference coverage policy versions without opening each database separately.

Named launch partners are AdventHealth, Baylor Scott and White Health, Boston Children's Hospital, Cedars-Sinai, HCA Healthcare, Memorial Sloan Kettering Cancer Center, and UCSF. The EHR integration requires a ChatGPT for Healthcare or eligible Enterprise account with a Business Associate Agreement in place; it is not available for individual accounts. Individual U.S. ChatGPT for Clinicians users can install the Healthcare Public Data plugin without an enterprise contract. Epic holds health records for more than 325 million patients, per TechCrunch citing Epic's own figures.

OpenAI also reported that physicians rated 99.1% of responses safe across 4,363 ratings spanning 27 clinical use cases, including pre-visit review, medication review, and handoff summaries, per the announcement. In a separate two-round evaluation, more than 93% of responses across each of five connected data sources tested were rated "good" accuracy or better.

Why it matters

Health-system IT teams evaluating AI adoption now have a concrete governance model to assess: read-only EHR access, BAA requirement, and enterprise-only EHR channel separated from the individual-clinician public data plugin. That separation defines rollout sequencing. A health system can offer the Healthcare Public Data plugin to individual clinicians today while negotiating the enterprise EHR contract on a longer timeline.

OpenAI noted that organizations with a BAA can use ChatGPT Work and Codex inside the same governed workspace, per the announcement. Clinical queries, administrative tasks, and software tooling all run under one access-control layer, which reduces the number of separate AI products a health system needs to govern separately. The physician safety ratings come from OpenAI's own evaluation program, so independent real-world error-rate data will be the next meaningful signal to watch.

What to watch next

Oracle Health (Cerner), Epic's main EHR competitor, has not announced an equivalent ChatGPT or Claude integration. Whether Oracle moves to match this or whether Anthropic partners with a competing EHR vendor will shape which AI products become the default inside clinical workflows. CMS and HHS have not yet issued formal guidance on HIPAA compliance for AI-assisted clinical documentation; regulatory clarity on that front would affect how quickly health systems can expand beyond pilot programs.

Sources