Google Cloud launched Gemini for Healthcare in preview on November 10, 2026, and its connections to Epic and Oracle Health are read-only. The agents pull a patient’s active conditions, medications and recent labs from the electronic health record and never write back to it. The same announcement says the agents orchestrate and execute multi-step workflows within guardrails, through connectors to eligibility, benefit verification and prior authorization services. The read-only promise covers the chart. It does not describe everything the agents can touch.
Gemini for Healthcare Ships With Read-Only Epic and Oracle Health Connectors
Gemini for Healthcare reaches Epic through the open.epic APIs and Oracle Health through its Ignite APIs for FHIR, according to Google’s press release. Google frames the read-only design as a way to protect clinical data integrity. NextGen Healthcare connectivity is in development and due in the coming months. Around the EHR links sit three more data services: InterSystems FHIR Server for repository management, Verato for matching patient identity across visits, and Keragon, which Google says extends connectivity to more than 400 healthcare applications. The product is built on the universal Gemini agent and sold to Gemini Enterprise customers. It ships with prebuilt skills for records synthesis, pre-appointment preparation, coverage and policy review, and payer benefit determination, and teams can write custom skills of their own.
The Payer Side of Gemini for Healthcare Connects to Systems That Run Workflows
The administrative connectors are where the agents move from reading to doing. Google lists lookups into CMS Coverage Data, the National Provider Identifier Registry, RxNorm, ICD-10 and the IMO Health Knowledge Graph for coding accuracy. Optum Network Eligibility and Coordination of Benefits is in limited release and testing for eligibility checks. Waystar and Cognizant Trizetto cover eligibility and payment records, GoodRx supplies prescription pricing, and Infinitus Relay automates benefit verification and supports prior authorization workflows. Further out, Suki provides ambient clinical documentation, UiPath automates chart summaries for utilization reviews and appeals, and the platform links to Google Workspace, Microsoft 365, ServiceNow and Salesforce’s Health Cloud. The release spells out the read-only boundary for EHRs only. It does not say which of the other connectors can submit, update or send.
Google Puts Google, Anthropic and OpenAI Models Under One Governance Layer
Google says applications, skills and connectors share a single governance framework that preserves an organization’s existing role-based access controls and adds audit logging and verifiable citations. Check Point handles perimeter defense and Cyera covers data and agent security. The platform supports models from Google, Anthropic and OpenAI, and a routing layer sends high-volume administrative work to fast, low-cost models while saving high-reasoning models for complex clinical and operational analysis. Thomas Kurian, Google Cloud’s chief executive, described the problem as “the immense friction of making that data actionable.” Launch customers named in the release are Blue Cross Blue Shield of Massachusetts, Hartford HealthCare, HCA Healthcare, Highmark Health, Northwell Health, Optum and Stanford School of Medicine, with the American Society of Clinical Oncology contributing guidelines content.
Why This Matters for Engineering Leadership
Keeping agents out of the system of record’s write path is a sound default, and Google made it a headline feature. The lesson for anyone evaluating healthcare agents, or agents in any regulated workflow, is that a read-only boundary is only as strong as its inventory. An agent that cannot edit a chart can still run a benefit verification, assemble a prior authorization or push a summary into an appeals queue, and those actions reach payers, patients and money. Before a pilot, ask the vendor for a table with one row per connector: whether the agent reads, writes or submits; whose identity the action runs under, the clinician’s or a service account; and how the audit log ties each action back to the person who approved it. Treat any connector without a clear answer as a write path. Then look at model routing. A platform that can send work to models from three different companies needs your data agreements to cover each one that will see patient information, and the routing rules should be something your team can read and pin. The read-only EHR line is worth copying into your own agent designs. Draw the same line, explicitly, around every other system the agent can reach.
FAQ
What is Gemini for Healthcare?
It is an agentic AI system from Google Cloud for healthcare providers and payers, built on the Gemini agent. Google launched it in preview on November 10, 2026.
Can Gemini for Healthcare write to Epic or Oracle Health records?
No. Google says its Epic and Oracle Health integrations are read-only and pull conditions, medications and recent labs without writing back to the EHR.
Which organizations are launch customers for Gemini for Healthcare?
Google named Blue Cross Blue Shield of Massachusetts, Hartford HealthCare, HCA Healthcare, Highmark Health, Northwell Health, Optum and Stanford School of Medicine, with ASCO contributing guidelines.
Which AI models does Gemini for Healthcare use?
Google says the platform supports models from Google, Anthropic and OpenAI, routing high-volume administrative tasks to fast low-cost models and complex analysis to high-reasoning models.
Does Gemini for Healthcare handle prior authorization?
Google says Infinitus Relay automates benefit verification and supports prior authorization workflows, with Waystar, Cognizant Trizetto and an Optum service covering eligibility checks.
What should engineering leaders ask before piloting healthcare agents?
Ask for a connector-by-connector list of which systems the agent can only read and which it can change, and which identity each action runs under.
