OpenAI is making its biggest push yet into clinical workflows, announcing that healthcare organizations can now connect Epic electronic health records directly to ChatGPT for Healthcare. The integration lets clinicians query authorized patient data conversationally — asking what changed since a patient’s last visit, which lab results need review, or what medication adjustments a specialist recommended. ChatGPT pulls from appointment notes, labs, medications, and specialist documentation, then summarizes developments and points back to the underlying chart entries. The company is also launching a Healthcare Public Data plugin that connects nine official sources, including ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed, and PubMed, giving teams structured access to trial eligibility criteria, medication identifiers, coverage policies, and provider records without jumping between separate databases.
OpenAI is backing the rollout with physician-led evaluations. The company says it works with hundreds of physicians across 60 countries, 49 languages, and 26 specialties, and those doctors have reviewed more than 700,000 model responses to date. For EHR-connected use cases specifically, physicians evaluated responses across 27 clinical scenarios — pre-visit review, clinical timelines, medication review, handoff summaries — and rated 99.1% of 4,363 responses as safe. In a separate evaluation covering five connected data sources, more than 93% of responses were rated “good” or better accuracy on nuanced clinical questions built on large U.S. healthcare datasets.
This is a meaningful escalation from OpenAI’s earlier healthcare positioning, which leaned heavily on ambient documentation and patient-facing tools. Native Epic integration puts ChatGPT inside the chart itself, not alongside it. The distinction matters: clinicians don’t have to leave their workflow to get AI assistance, which has been the adoption killer for most clinical AI tools. That said, safety ratings from vendor-sponsored evaluations have a mixed track record in this industry. A 99.1% safe rating sounds reassuring until you consider the denominator — 4,363 responses across 27 use cases is a thin slice of what a real deployment will encounter. The failure modes in EHR data are notoriously edge-case heavy: duplicate meds, conflicting notes, outdated problem lists. General evaluations rarely surface those.
The two-way integration design is worth watching. Bringing EHR context into ChatGPT lets clinicians interrogate the record conversationally, while embedding ChatGPT into the EHR layout keeps AI inside the governed clinical workspace. That dual approach acknowledges a hard truth in health IT: clinicians won’t adopt tools that force them to switch contexts. But the real test comes when a physician asks about a patient with a messy, contradictory chart — the kind of record that makes up a substantial share of real-world EHR data. OpenAI’s safety numbers are a starting point, not proof. The deployments that follow will show whether the integration holds up when the chart fights back.
