Google launches Gemini for government services in preview for US agencies

What Google announced

Google Public Sector, the Google Cloud unit that sells to governments, has launched Gemini for government services, a packaged version of its Gemini agent platform built for federal, state and local agencies in the United States. The company announced it on 20 October 2026 at its Google Public Sector Summit in Washington.

The product is in preview. Google says it is already being used by select agencies, naming two: the State of Missouri and the US Department of Health and Human Services. Those two have been testing it over the past few weeks, Chris Hein, field chief technology officer at Google Public Sector, told Nextgov/FCW.

It is the third industry package of its kind, after Gemini for legal and Gemini for financial services. Each one bundles agents, ready-made skills, data connectors and model tuning for one sector on top of the same governed Gemini platform that Google Cloud has been pitching to companies, including the Gemini agent it named its universal workplace agent earlier this month.

Google Cloud chief executive Thomas Kurian framed the launch around trust. “To deploy agentic capabilities in the public sector requires non-negotiable standards for security, confidentiality, and data sovereignty,” he said in the announcement.

Skills, connectors and partner agents

Google describes four parts. The first is a set of skills: pre-loaded instructions that guide an agent through an administrative job, such as digesting federal and state regulations, working through acquisition and procurement steps, or helping staff evaluate vendor qualifications.

Hein told Nextgov/FCW that a skill works as a set of rules telling Gemini how to carry out a task without going off script. Staff call one up from the Gemini prompt bar, the way a user might invoke a saved command, rather than writing a long prompt each time.

The second part is connectors. Google lists Google Workspace, Microsoft SharePoint, Outlook 365, Salesforce and ServiceNow, plus a connector built by the digital services firm BLEN that searches the Electronic Code of Federal Regulations, regulations.gov and the Federal Register. Custom connectors reach Oracle enterprise software such as Oracle Fusion Cloud Applications.

The third part is third-party agents and consultants. Google names Accenture Federal Services, Clutch, Dark Wolf, Daston, Deloitte, Egen, Game Plan Tech, Knexus, Onix, Resultant, SAIC, Slalom and SZNS Solutions. Hein said Google will prescreen partner agents, and that agency IT leaders, not individual employees, switch them on through an administrative portal. Some agents are available now, but most are not yet, he said.

The fourth part is the platform itself, which Google describes as a central control plane for IT and risk teams, with risk management, audit logging and governance built in.

The jobs Google is targeting

Google picked three kinds of work to show the product off. The first is reading rules. Staff can query federal statutes, state administrative codes and internal policy memos and get answers with page and paragraph citations, so they can check the source before acting on the answer.

The second is procurement. Google says the agents can cross-check multi-volume vendor bids against a request for proposals and the Federal Acquisition Regulation, flag where a bidder falls short, score submissions against a standard rubric and draft a justification that auditors can follow.

The third is work with the public. Google says the system can classify documents citizens submit, such as pay stubs, W-2 tax forms and leases, and “mathematically validate” them, and can rewrite dense regulatory language into plain notices that help agencies meet the Plain Writing Act of 2010. It also lists redacting records released under the Freedom of Information Act.

Missouri, one of the two early users, put the goal in practical terms. “Our citizens expect the government to move at the speed of modern life, not the speed of paperwork,” said Tim Marczewski, director of AI and innovation in the state's Office of Administration.

How it looked in the demo

Nextgov/FCW described a video demonstration in which a compliance analyst working for a state chief information security officer used a skill called the policy and statute assistant to analyse federal cybersecurity rules.

The analyst then asked Gemini to draft a ServiceNow incident ticket to bring the relevant contracts into line with those rules. The ticket could be edited before it was sent, which shows where Google is placing the human check: the agent drafts the change, and a person decides whether to file it.

That pattern, an agent reading rules and then acting inside a system of record, is what separates this from the chat assistants agencies already have. It is also where mistakes carry more weight, because a wrong reading of a regulation can now turn into a filed ticket or a scored bid rather than just a bad answer on screen.

Data handling and control

Google says the service connects directly to agencies' core systems and inherits their existing identity and access controls, so an employee should only reach files they could already open. It says answers are drawn from official regulations and the agency's own files, with links back to the exact source.

Agency and citizen data, prompts and outputs stay inside the organisation's cloud perimeter, according to Google, and base models are never trained on customer data without the customer's consent. Sensitive records are kept isolated in line with each agency's security rules.

Those promises matter in a market where the rules are getting stricter. The General Services Administration has added a contract clause on how LLM systems handle government data, and the federal government has just ordered agencies to remove DeepSeek and tools built on it. Vendors now compete on where data goes as much as on model quality.

When the product reaches general availability, Hein said, agency IT leaders will be able to choose which connectors to enable and for which groups of users.

Price and the OneGov deal

Google did not publish a price for Gemini for government services. Nextgov/FCW reported that federal agencies can get Gemini for $0.47 per agency per year through 15 November under a recently extended OneGov deal.

That deal dates from August 2025, when the General Services Administration announced a OneGov agreement for a Gemini for Government suite at $0.47 per agency, then described as valid through 2026. GSA said at the time that it covered enterprise search, NotebookLM, video and image generation, pre-packaged agents and tools for staff to build their own agents.

It is not clear whether the new skills, connectors and partner agents fall under that discounted agreement, cost extra, or will be priced once the preview ends. State and local governments are not covered by the federal OneGov terms at all.

What is still missing

This is a preview. Hein said Google is taking on more early adopters to gather feedback, and that general availability is expected later this year, though there is no target date.

Google's release says the service “will meet” the highest security and compliance standards required by US agencies, in the future tense, and does not name a specific authorisation level for the new product. Agencies handling sensitive data will want that spelled out.

The accuracy claims are untested in public. Google has not published error rates for its FOIA redaction, bid scoring or document validation, and has not said what “mathematically validate” means for a pay stub. The models underneath, such as Gemini 4 Argon, are capable, but a missed redaction or a wrongly scored bid has legal consequences.

Google's bet is that packaged, rule-following skills plus access controls agencies already trust will move AI from pilots into daily casework. The preview at HHS and Missouri is the first real test of that.