OpenAI has connected ChatGPT for Healthcare directly to Epic Systems, the dominant electronic health record platform used across the United States, giving clinicians a new way to pull patient charts, lab results and medication histories into the chatbot before and during appointments. 

 

The integration, announced September 1, 2026, marks OpenAI's most concrete step yet toward embedding its consumer-facing AI inside the clinical workflows that hospitals and physician groups already run on every day.

 

Epic's software underpins medical records for more than 325 million patients in the U.S., making it the largest single gateway OpenAI has connected ChatGPT to inside healthcare. The company built two distinct paths into that data. 

 

Clinicians can pull authorized patient information — appointment notes, lab results, medications and specialist documentation — into ChatGPT ahead of a visit to get a quick summary or a timeline of what has changed since the last appointment. Alternatively, in supported deployments, clinicians can call up ChatGPT directly inside Epic's own workflow without leaving the patient's chart.

 

Crucially, OpenAI says the connection is read-only. ChatGPT can retrieve and summarize what is already in the record, but it cannot write anything back into a patient's chart, a design choice clearly aimed at limiting the AI's ability to introduce errors into the medical record itself even as it gains broader access to what's inside it.

 

UCSF Health is serving as a pilot site for the rollout, with AdventHealth also named as a health-system partner. UCSF Health CEO Suresh Gunasekaran said the technology "has the potential to reduce time spent synthesizing data" by bringing relevant patient information together more quickly and comprehensively — a pitch aimed squarely at the administrative burden that has become one of the most cited sources of physician burnout.

 

Alongside the Epic connection, OpenAI introduced a Healthcare Public Data plugin that links ChatGPT to a set of official public health databases, including ClinicalTrials.gov, CMS coverage data, RxNorm, DailyMed and PubMed. 

 

The plugin is designed to let clinicians compare clinical-trial eligibility criteria, medication labeling and insurance coverage policies without searching each database separately — folding reference lookups that currently take multiple browser tabs into a single query.

 

OpenAI paired the launch with compliance infrastructure aimed at health systems' legal and IT departments: organizations with a signed Business Associate Agreement can run ChatGPT Work, Codex, apps and connectors under HIPAA-compliant workflows that include role-based access controls, single sign-on and audit logging. 

 

That framework is what allows a hospital to plug ChatGPT into a system holding protected health information without violating federal privacy law, at least on paper — the compliance burden and liability exposure still sit with the health system that adopts it.

 

On accuracy, OpenAI published internal validation figures rather than independent third-party results: physicians reviewed 4,363 ChatGPT responses across 27 clinical use cases and rated 99.1% of them safe, with accuracy above 93% across the five data sources the system draws on. 

 

The company says it works with more than 700 physician advisors spanning 60 countries and 26 medical specialties to shape how the product behaves in clinical settings.

 

The Epic integration extends a healthcare push OpenAI has been building since the start of the year. ChatGPT for Healthcare was first announced in January 2026, followed by ChatGPT for Clinicians in April for documentation and medical research tasks, and a consumer-facing ChatGPT Health rollout to all U.S. users in July. The underlying models are versions of GPT-5. 

 

OpenAI says users now send roughly 300 million health-related queries to ChatGPT every week, underscoring both the scale of consumer interest and the stakes of getting clinical answers right.

 

Those stakes are not hypothetical. OpenAI continues to maintain publicly that ChatGPT "is not suitable for diagnosis or treatment," language that has taken on more weight amid pending litigation. 

 

A Florida pastor sued OpenAI in July 2026 alleging the chatbot gave him medical advice that nearly proved fatal, and in May 2026 family members of another user sued over what they described as a wrongful medication dosage recommendation from the chatbot. 

 

Those cases involve consumer use of ChatGPT rather than the new clinician-facing Epic integration, but they illustrate the liability backdrop against which OpenAI is now asking hospitals to route real patient charts through the same underlying models.

 

For OpenAI, the Epic partnership is a significant enterprise and distribution win: Epic's near-ubiquity in large U.S. health systems means integrating with it, rather than building bespoke connections hospital by hospital, is the fastest route to clinical-workflow scale. 

 

It also puts OpenAI in more direct competition with a growing field of AI tools purpose-built for clinical documentation and chart review that have been selling into health systems on the strength of narrower, more specialized workflows.

 

Whether hospitals embrace the integration at scale will likely hinge less on OpenAI's internal safety statistics than on how health systems' own compliance, legal and clinical-informatics teams weigh the benefits of faster chart synthesis against the risks of routing protected health data through a general-purpose consumer AI model — even one restricted to reading, not writing, the record. 

 

UCSF Health's pilot, and how quickly other Epic-run systems follow AdventHealth in adopting the connector, will be an early signal of how that calculation is playing out.