On September 1, 2026, OpenAI announced two data-connection updates for ChatGPT for Healthcare: an Epic electronic health record integration and a Healthcare Public Data plugin. The first brings authorized patient context into ChatGPT; the second puts nine official healthcare sources into one queryable workflow. The goal is not to replace a hospital system with a chatbot, but to put approved context inside a governed workspace.
After a healthcare organization authorizes it, the Epic integration can provide patient data and relevant fields to ChatGPT for Healthcare. OpenAI also describes workflows that use ChatGPT inside the EHR. The Healthcare Public Data plugin covers official sources including ClinicalTrials.gov, CMS Coverage, RxNorm, DailyMed, and PubMed, allowing users to compare trial criteria, medication identifiers, coverage, provider, and patient records. Versioned sources and identifiers make this more traceable than asking a general-purpose model without a defined evidence base.
Governance is part of the product positioning. The governed ChatGPT for Healthcare workspace includes RBAC, SSO, and audit logs. With the applicable agreement and a BAA, organizations can use it in workflows designed to support HIPAA compliance. The Epic integration is not for individual accounts, while the public-data plugin is intended for eligible US ChatGPT for Clinicians users. Access is therefore tied to the account, institution, and data-processing arrangement.
The announcement cites two evaluation sets. OpenAI says physicians rated 27 clinical use cases 4,363 times, with 99.1% of responses rated safe. A separate evaluation across five sources found more than 93% of results good or better. These are OpenAI-published internal or partner evaluations, not independent medical audits, and they are not a guarantee of accuracy or clinical safety. Real deployments still face missing data, version differences, permission mismatches, and model misinterpretation.
The workflow value is mainly reducing context switching, not automating clinical decisions. Within approved boundaries, a clinician can retrieve a chart summary, compare trial criteria, check a medication name, or read public research before returning the work to the existing professional review process. Organizations should specify which data may enter a prompt, which answers must cite primary sources, when a physician must review, and how every query or change is retained in the audit log.
There are clear limits. An official source does not mean every patient record is complete. EHR access does not mean the model understands the full clinical context. A BAA does not automatically make every workflow compliant. ChatGPT should not diagnose, prescribe, or direct emergency treatment without qualified professional review. For high-risk actions, a safer pattern is for AI to organize evidence and surface uncertainty, then have a clinician decide and document the outcome.
By placing Epic, public data, enterprise identity, and audit tools in one announcement, OpenAI is signaling that healthcare AI competition is moving beyond answer quality toward controlled work inside existing systems. The meaningful test will be whether authorization, data versions, citations, human approval, exception handling, and post-use traceability connect reliably in practice.
