ChatGPT Health connects clinicians to Epic patient records

Inside a patient chart, a clinician can now ask ChatGPT Health to organize appointment notes, laboratory results, medications and specialist documentation imported from Epic’s electronic health record, or EHR—a digital patient record system. Epic’s system holds data for over 325 million patients, according to OpenAI, and in certain deployments ChatGPT appears directly within the existing EHR workflow.

The practical mechanism is simple: the AI reads the record, summarizes it, highlights changes in the patient’s history and helps prepare for a future appointment. It can also build a clinical timeline or conduct a pre-visit review. The boundary is equally clear: the integration is read-only, and ChatGPT does not write anything back into the chart.

OpenAI is adding a Healthcare Public Data plugin alongside the Epic connection. It can retrieve information from ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed and PubMed, helping healthcare workers synthesize trial eligibility criteria, medication identifiers, coverage policy versions and provider records. Organizations with a Business Associate Agreement can also use ChatGPT Work, Codex, apps and connectors for compliant workflows; the agreement is context for the deployment, not a new clinical capability.

For clinicians, the integration can help consolidate information before an appointment, during a review or at handoff; in some deployments, they can use it without leaving the patient chart. OpenAI says it gathered over 4,300 responses from physicians across 27 clinical use cases, with 99.1% judged safe. That is the company’s own assessment, not an independent clinical validation, and even a small number of unsafe answers can harm patients.

The caution is not theoretical. OpenAI has said ChatGPT is not suitable for diagnosis or treatment, while lawsuits in Florida and elsewhere allege that ChatGPT gave harmful medical or dosage advice. Epic integration can make records easier to navigate; it does not turn the model into a clinician or eliminate the risk of unsafe answers.