OpenAI and its health-system partners see the Epic integration as a way to cut the clerical burden on clinicians; the company’s own rollout, however, arrives under the shadow of lawsuits and its insistence that the tool is not for diagnosis or treatment.

Last month, OpenAI expanded ChatGPT for health to U.S. consumers, saying users were already putting roughly 300 million health-related questions to the chatbot each week. Days before the latest enterprise announcement, a Florida pastor sued, alleging ChatGPT had offered a near-fatal recommendation; another lawsuit filed in May blamed the system for harmful dosage advice.¹

Now OpenAI is moving deeper into clinical infrastructure. Healthcare organizations can connect authorized Epic records to ChatGPT for Healthcare, allowing staff to pull together appointment notes, lab results, medications and specialist documentation. In supported deployments, the assistant can sit inside the electronic health-record workflow — but its access is read-only, meaning it cannot write back into a patient chart.²

The company frames the change as a decision-support tool, not an automated clinician. It also added a Healthcare Public Data plugin linking nine official sources, including PubMed, DailyMed, ClinicalTrials.gov and CMS Coverage, so teams can check research, drug labels, trials and coverage rules without hopping among databases.²

OpenAI says physicians rated 99.1% of 4,363 responses across 27 clinical use cases as safe. But the remaining fraction carries outsized weight when the output concerns a patient. UCSF Health chief executive Suresh Gunasekaran called the pilot’s promise a chance to “reduce time spent synthesizing data and give clinicians more time with patients,” while stressing that frontline teams would validate the capabilities in practice.²

That is the central bargain: faster access to a sprawling medical record, with humans still responsible for deciding what the machine’s summary means.