Google Cloud has put Gemini for healthcare into preview in the United States, extending its Gemini agent into hospitals and health plans. The package connects clinical records, claims and coverage systems, and runs multi-step work with human oversight. It is available now in preview for Gemini Enterprise customers.
What the Package Does
The product is built around four parts. Skills written by domain experts teach Gemini to run tasks, from medical records synthesis to coverage review. Connectors link it to the systems hospitals already run. An open partner ecosystem extends it, and a governed platform sets the rules for every agent interaction.
The announcement names 27 connectors. Epic and Oracle Health records open read-only, with no write-back to the chart, so the medical record stays intact. Claims, eligibility and pricing tools come from Optum, TriZetto and GoodRx, while CMS coverage rules and drug vocabularies ground each answer in published policy.
At the point of care, Gemini builds a timeline from fragmented records, separating active problems from resolved ones and flagging medication discrepancies. It prepares charts before visits and assembles authorization packets against payer policy, drafting appeals from the record when a denial lands.
For health plans, the agent checks requests against medical policy and organizes findings for clinical reviewers, with each point traceable to the submitted documents. It also explains benefits from the member’s own plan documents, estimating out-of-pocket costs, tracking deductibles and verifying in-network status.
Google says clinical data, prompts and generated outputs stay inside each organization’s governed environment and are never used to train its foundation models. The system is built to support HIPAA compliance, with audit logging and verifiable citations so reviewers can check where each answer came from.
The platform also offers open model choice, pairing Google’s frontier models with partner models for complex reasoning while cheaper models handle routine steps. Sixteen integrators, including Accenture, Deloitte and Cognizant, will customize and deploy it, alongside agents such as an ASCO oncology guideline assistant.
Google developed the package with HCA Healthcare, Highmark Health, Northwell Health, Optum and Blue Cross Blue Shield of Massachusetts. The company says clinicians spend nearly 28 hours a week reconciling information, a burden the agent is meant to cut — though that figure is Google’s own, with no study attached.
Our Read
The read-only design is the whole pitch. Earlier health AI stumbled on trust: hospitals would not let a model touch the record. By promising it can read everything and change nothing, Google answers the safety objection first and leaves clinicians in charge of sign-off.
The open model choice is a quieter move. Google is telling health systems they need not bet on one lab’s models, which undercuts rivals that sell a single stack. The hospital, not the vendor, sets the rules for every agent.
Rivals are chasing the same paperwork. Microsoft’s Nuance tools and Amazon’s HealthScribe already target clinical documentation, and Apple is now studying how people use AI for health answers. The real test is not the launch list but whether hospitals trust an agent with patient data at scale.
Why It Matters
Administrative work is the bottleneck in healthcare, and it is where AI can help soonest. If an agent can assemble a packet in minutes, clinicians get time back for patients. But this is a preview, not a result: no patient outcomes exist, and no price or availability date was published.
Background
Health AI has drawn both investment and scrutiny this year. Apple opened a 200,000-person study of how people use AI for health questions, while U.S. regulators dropped a rule requiring AI model cards in certified software. Google’s own Gemini skills reached Workspace days before this launch.
Source: Google Cloud blog


0 Comments