--- title: "Gemini for Healthcare Reads the Chart. Whose Infrastructure Runs the Agent?" slug: "gemini-for-healthcare-google-cloud-agents-whose-infrastructure" author: "Blanca Amigot" date: "2026-11-12 10:00:00" category: "Premium" topics: "Gemini for healthcare, Google Cloud healthcare AI, healthcare AI agents, prior authorization AI, Epic integration, Oracle Health, read-only EHR connectors, PHI, HIPAA, health system AI architecture, own the code" summary: "Google Cloud put Gemini for healthcare into U.S. preview on November 10, 2026, with read-only Epic and Oracle Health connectors and a promise that clinical data stays in the customer's governed environment on Google Cloud. The question left for a health system is where the agent runs and who owns it." banner: "" thumbnail: "" linkedin: | On November 10, 2026, Google Cloud put Gemini for healthcare into preview in the U.S. It gives providers and payers agents that prepare charts, assemble prior authorization packets against payer policy, and draft appeals grounded in the record. Two design choices deserve credit. The Epic and Oracle Health connectors are read-only, with no write-back to the chart. And Google says clinical data, prompts and outputs stay within the organization's governed environment and are never used to train its foundation models. So the privacy question is mostly answered. The architecture question is not. Google's announcement says the product runs on Google Cloud's infrastructure, and it lists no pricing. For a health system that runs on another cloud or its own data centers, that is the decision: rent an agent layer inside one vendor's cloud, or own the agent layer and run it where the patient data already lives. #iblai #HealthcareAI #PriorAuthorization #AIAgents #HealthIT --- ## The Short Answer **Google Cloud's Gemini for healthcare, in U.S. preview since November 10, 2026, gives providers and payers agents with read-only Epic and Oracle Health connectors, and keeps their data in a governed environment on Google Cloud. The remaining question is location and ownership: on ibl.ai you own all the code and the data, and the agents run where the health system chooses.** Google's announcement is [Introducing Gemini for healthcare](https://cloud.google.com/blog/products/ai-machine-learning/gemini-for-healthcare), published on the Google Cloud blog on November 10, 2026. ## What did Google Cloud announce with Gemini for healthcare? On November 10, 2026, Google Cloud [introduced Gemini for healthcare](https://cloud.google.com/blog/products/ai-machine-learning/gemini-for-healthcare), an extension of its Gemini agent for health systems and health plans. It is available in preview in the U.S. For providers, it prepares a patient's chart before a visit and assembles prior authorization packets against payer policy before submission. For clinical authorization and appeals, it flags documentation gaps and, where the record supports it, drafts an appeal grounded in the chart and the plan's policy. The announcement names 27 connectors, among them Epic, Oracle Health, InterSystems FHIR Server, CMS Coverage, the NPPES NPI Registry, ICD-10 codes, RxNorm, Optum eligibility tools, Infinitus Relay for payer phone calls, Salesforce, ServiceNow and Microsoft 365, plus 16 systems integrators. ## Does Gemini for healthcare write back to the Epic or Oracle Health chart? No. Google describes the Epic connector as giving authorized users "read-only access to patient conditions, medications, labs, clinical notes, and specialist documentation," with "no write-back to the underlying chart." The Oracle Health connector likewise enforces "zero write-back." That is a sound default for a first release. An agent that cannot write to the system of record cannot corrupt it, and the human reviewer stays in the loop. [Futurum's analysis](https://futurumgroup.com/insights/gemini-for-healthcare-reads-the-ehr/) notes that whether the EHR vendors will later allow governed write-back is an open question. OpenAI made the same read-only choice when it connected ChatGPT for Healthcare to Epic in September; we covered that launch in [OpenAI Wired ChatGPT Into Epic. Where Does the PHI Go?](/blog/openai-chatgpt-epic-ehr-platform-era-healthcare) ## Where does patient data go in Gemini for healthcare? Not as far as the launch slogans suggest. Google states that "clinical data, prompts, and generated outputs remain within your organization's governed environment and are never used to train or fine-tune Google's foundation models," and that the product is "built to support HIPAA compliance." It also states that Gemini for healthcare "runs on Google Cloud's secure infrastructure." So the governed environment is the customer's own space inside Google Cloud. The announcement does not describe running the agent on another cloud or in a health system's own data center. For a health system already standardized on Google Cloud, that may be exactly right. For one whose EHR, data warehouse and identity stack sit on Azure, AWS or on-premises hardware, the connectors can reach those systems where they are, but the agent layer itself runs in Google Cloud. ## Is Gemini for healthcare locked to Google's models? No, and it is worth being precise here. Google says the agent "supports model choice," so organizations can pair frontier models "from Google and leading partners" for clinical reasoning with cheaper models for routine administrative steps. The announcement does not name the partner models. Model choice inside a vendor's agent is not the same as owning the agent. The connectors, the orchestration, the governance layer and the prompts that encode a health system's prior authorization workflow live in the product. If the terms or the price change, that workflow does not move with you. ## What does Gemini for healthcare cost? Google has not said. The announcement lists no pricing, and [Futurum](https://futurumgroup.com/insights/gemini-for-healthcare-reads-the-ehr/) observes that it is not stated whether pricing will be per seat, consumption-based or outcome-based. That matters for revenue-cycle work in particular, because the workload scales with claim volume, not with headcount. We worked the per-letter arithmetic in [what AI prior authorization actually costs](/blog/what-ai-prior-authorization-actually-costs-2026), and why billing is the first place to deploy in [Healthcare AI Should Start in the Billing Office](/blog/healthcare-ai-revenue-cycle-prior-authorization-denials). ## How should a health system compare a cloud vendor's healthcare agent with owning the agent layer? Ask the same five questions of every option, Gemini for healthcare included:
Question Gemini for healthcare (per Google, Nov 10, 2026) On ibl.ai
Where does the agent run? Google Cloud infrastructure Your cloud, on-premises, GovCloud or air-gapped
Who owns the agent code? Google (a managed product) The health system, under a perpetual license
Can it write to the chart? No, read-only connectors Role-scoped, read-only permissions, set per agent through MCP connectors
Which models? Google and unnamed partner models Any LLM, including open-weight models run locally
How is it priced? Not disclosed in preview Usage-based, no per-seat pricing
None of this makes the managed option wrong. A health system that wants a vendor to run the agent layer, and already runs on Google Cloud, gets a coherent product with strong defaults. The table is for the systems that want the layer to be theirs. ## How does ibl.ai run healthcare agents inside a health system's own perimeter? On ibl.ai you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing, so you can deploy anywhere, from your own cloud to a fully air-gapped network. The agents reach the EHR and payer systems in place, through connectors the health system controls. Every run leaves a record the health system keeps: [transcripts](/docs/os/analytics/transcripts) and [audit](/docs/os/analytics/audit) logs, plus [evaluations](/docs/os/agent-settings/evals) to score a prior authorization agent on the system's own cases before a model change goes live. These are existing ibl.ai capabilities. Syracuse University runs ibl.ai this way at ai.syracuse.edu, as told in our [case studies](/case-studies), on its own cloud and behind its own firewall. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on. [Book a 30-minute demo](https://cal.com/iblai/30min) and we will scope a prior authorization agent on your own infrastructure. *Sources: launch date, preview status, capabilities, the 27 named connectors and 16 integrators, the read-only and data-handling statements, model choice and the Google Cloud infrastructure statement from Google Cloud's [Introducing Gemini for healthcare](https://cloud.google.com/blog/products/ai-machine-learning/gemini-for-healthcare) (November 10, 2026); pricing and write-back analysis from [Futurum](https://futurumgroup.com/insights/gemini-for-healthcare-reads-the-ehr/). The brief said "every patient chart flows through Google's infrastructure." Google states that clinical data, prompts and outputs stay within the customer's governed environment and are not used for training; that environment is on Google Cloud, which is the narrower and accurate point. The product is a U.S. preview, not a general release, and it assembles prior authorization packets rather than submitting claims.* ## Why does owning the AI stack matter? **ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.** - **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours. - **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform. - **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ. - **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.