Gemini for government services entered preview on October 20, 2026. In an official blog post, Google Cloud CEO Thomas Kurian announced a version of the Gemini agent rebuilt for U.S. federal, state, and local agencies, combining five parts: skills written for government work, connectors to the systems agencies already run, partner-built agents, a governed control plane, and an open partner ecosystem. It is preview-only for now, with no general-availability date.
It starts with the paperwork
The skill list reads like a civil servant's week: policy and regulatory research with answers cited to the page and paragraph of the CFR, the Federal Register, and state codes; discovery and batch redaction for freedom-of-information requests, with each proposed redaction labeled by its statutory exemption and audit-ready copies produced only after a reviewer confirms; vendor bid evaluation scored against procurement requirements with citations; grant expense audits that check invoices and ledgers line by line against federal guidance; document intake validation that catches missing signatures before an application joins the queue; and plain-language rewrites of policy memos for the public.
The real pitch is not moving the data
Connectors query agency systems in place, so nothing is migrated, and access follows the identity and record-level permissions each system already enforces. Federal sources include the Federal Register, regulations.gov, and the electronic CFR; systems of record include Oracle E-Business Suite and PeopleSoft, Salesforce, and ServiceNow; collaboration spans both Google Workspace and Microsoft 365 with SharePoint. Partner agents from SZNS Solutions for records discovery and redaction run under the same governance, with a person approving each consequential step.
Governance comes before features
The service can run inside a FedRAMP High data boundary with U.S. data residency and personnel controls, and Google commits that agency records, custom agents, and model outputs stay private and are never used to train or fine-tune its foundation models without permission. The State of Missouri and the U.S. Department of Health and Human Services helped develop it, and a Missouri official put the motivation bluntly: citizens expect government to move at the speed of modern life, not the speed of paperwork.
For public-sector teams elsewhere, the lesson is the delivery model, not the model itself: no rip-and-replace, no migration, citations and human approval as defaults. Government AI competition is shifting from whose model is strongest to who can plug into legacy systems and state the accountability boundaries clearly. How much of this survives real agency workflows will depend on preview feedback.
