ChatGPT Just Plugged Into 325 Million Patient Records. Seven Hospital Systems Are Already Using It.

On September 1, 2026, OpenAI announced something that would have seemed like a regulatory impossibility two years ago: ChatGPT for Healthcare now connects directly to Epic's electronic health record system, the software that manages care for more than 325 million patients across the United States. Seven major health systems — AdventHealth, Baylor Scott & White Health, Boston Children's Hospital, Cedars-Sinai, HCA Healthcare, Memorial Sloan Kettering Cancer Center, and UCSF Health — are already live.

The integration is read-only. ChatGPT cannot write to a chart, place orders, or modify clinical documentation. But within those boundaries, what it can do is significant: a physician can pull a patient's appointment notes, laboratory results, medication history, and specialist documentation into a single conversation and ask ChatGPT to make sense of it — in seconds, before walking into an exam room.

Two Ways In

The integration ships in two deployment modes. In the first, clinicians stay inside ChatGPT for Healthcare and pull authorized Epic data into the workspace — treating the EHR as a source the AI can query rather than a screen they have to context-switch between. In the second, ChatGPT embeds directly inside Epic's own interface, so physicians never leave the chart they're already working in.

Suresh Gunasekaran, president and CEO of UCSF Health, described what that looks like in practice: the AI helps clinicians "understand what has changed and what matters most across a complex patient record." For a subspecialist seeing a returning oncology patient with a 47-page chart, that is not a minor convenience.

The Safety Study

OpenAI published results from a pre-launch safety evaluation covering 27 distinct clinical scenarios — pre-visit summaries, medication reconciliation, clinical timelines, and handoff notes among them. Physicians rated 4,363 individual AI responses. The safe-response rate across all scenarios: 99.1%.

That figure sounds reassuring until you do the math on scale. Epic's network serves hundreds of millions of patients. Even a 0.9% error rate applied to a fraction of that user base represents an enormous absolute number of potentially dangerous outputs. OpenAI acknowledged this directly: "AI is not suitable for diagnosis or treatment," the company said in its launch materials, noting that "even a few unsafe answers can lead to harmful results for humans." At least two lawsuits against OpenAI reference harmful medical advice from ChatGPT in an earlier context.

The company's answer to that concern is structural rather than statistical: read-only access prevents the AI from acting on bad conclusions, HIPAA compliance is enforced via Business Associate Agreements with each health system, and every session carries role-based access controls, single sign-on, and full audit logs.

The Public Data Layer

Running alongside the EHR integration is a new Healthcare Public Data plugin that connects ChatGPT for Healthcare to nine authoritative external datasets: PubMed, ClinicalTrials.gov, DailyMed, RxNorm, and CMS Coverage, among others. The intent is to let the AI ground clinical reasoning not just in a patient's chart but in current medical literature, active trials, approved drug labeling, and federal coverage policy — all in a single query.

This is the more quietly consequential part of the announcement. A clinician asking about a rare drug interaction can now get an answer sourced from DailyMed's regulatory drug labels and PubMed's indexed literature simultaneously, within the same interface they use for patient records.

From Zero to 300 Million

ChatGPT Health launched in January 2026 as a controlled product for health system deployments. By July 2026 it had expanded to all US adults 18 and older. By that same month it was fielding 300 million health-related queries per week — a usage figure that suggests the product had already become a de facto medical reference layer for a significant portion of the US population before a single EHR was ever connected.

The Epic integration changes the nature of what ChatGPT Health is. Before September 1, it was a sophisticated question-answering system that happened to be used by clinicians for research and reference. Now it is a system that can see a specific patient's data, reason over it, and present structured conclusions to the physician managing that patient's care.

The Structural Question

What this integration does not yet address — and what will likely dominate the next phase of debate — is where clinical liability actually falls when a physician acts on an AI-generated chart summary. Epic's system is the authoritative record. ChatGPT's synthesis of it is a derived output, built on a 99.1% safety rating that was established in controlled evaluation scenarios, not in the full chaos of emergency medicine, rare disease, or polypharmacy edge cases that define real clinical practice.

For now, OpenAI has drawn a careful perimeter: the AI reads, it does not write. Seven hospital systems have decided that is enough to start. The other 2,600-plus US hospitals that run Epic have a decision ahead of them.

Sources

TechCrunch — ChatGPT Health adds Epic integration for clinicians to import patient data: https://techcrunch.com/2026/09/01/chatgpt-health-adds-epic-integration-for-clinicians-to-import-patient-data/

Unite.AI — OpenAI Connects Epic Health Records and Public Data to ChatGPT: https://www.unite.ai/openai-connects-epic-health-records-and-public-data-to-chatgpt/

CryptoBriefing — OpenAI Epic integration brings ChatGPT to health patient data: https://cryptobriefing.com/openai-epic-integration-chatgpt-health-patient-data/

PYMNTS — OpenAI Brings Epic Health Records to ChatGPT for Clinicians: https://www.pymnts.com/news/artificial-intelligence/2026/openai-brings-epic-health-records-to-chatgpt-for-clinicians/

StockPil — OpenAI ChatGPT Health Epic EHR integration: https://stockpil.com/openai-chatgpt-health-epic-ehr-integration