AI for government
AI for local government and state agencies, approved, built and run on Microsoft
We build and run AI for cities, counties, utilities, school districts and state agencies: inside your own Microsoft 365, with the model that fits the job, and governed from the first day.
On this page
Short answer
Communication Square builds and runs AI for public agencies on Microsoft 365 and Azure: AI policy and governance, records and document assistants, phone, chat and email assistants, and agents for high-volume casework. We work inside your own Microsoft 365, use OpenAI, Anthropic or open-source models hosted in Microsoft Foundry on Azure (never trained on your data), and put the controls in place before anything is switched on.
Why now
AI is the top priority in state IT, and ungoverned use is already costly
How it works
Every AI for local government engagement runs the same four steps
- Planning meetingA short call about what an AI assistant would find in your own Microsoft 365.
- Policy and planAn AI use policy and the controls each use needs, aligned to the NIST AI RMF.
- BuildThe first use built in your own Microsoft 365 with approval rules, logging and human review.
- Run and measureWe operate each agent and report what it does, every month.
Our promise: nothing goes live until you have signed off its controls, and every agent ships with a named owner, approval rules, an audit log and an off switch.
Governance and data handling
Where your data goes, and who stays in control
- Inside your own Microsoft 365. We build in your Microsoft 365 and Azure environment, using the identity and permissions you already govern.
- Exposure fixed first. An AI assistant sees whatever its user can see, so oversharing and unlabeled sensitive content are dealt with before anything is switched on.
- Hosted in Microsoft Foundry on Azure. We use the model that fits the job: OpenAI, Anthropic or open-source models, deployed in Microsoft Foundry on Azure. Your data is never used to train them (see Microsoft’s data terms for Foundry). Anthropic operates its Claude models on Azure under its own commercial terms, and we show you which model and which terms apply before work starts.
- People decide. Approval rules, audit trails and human review wherever an output affects a resident, a record or a decision.
Proof
Built on work agencies already trust us with
Typical investment
What a first engagement costs
Book a short meeting and we will give you the firm number. Managed AI operations run $2.7K–$10.5K a month.
Common questions
AI in local government: questions agencies ask
How can local governments use AI?
The uses that pay back first are the high-volume ones: answering resident calls and chats, finding records for public records requests, drafting recurring reports and checking permit and intake applications for completeness. Each starts with a policy and the controls that go with it; see AI policy and governance.
Can government employees use ChatGPT on government computers?
That is each agency’s decision, and many now answer it in an AI use policy. Blocking public tools tends to push use onto personal devices, so most agencies do better with an approved assistant inside their own Microsoft 365, clear rules on what data may go into a prompt and DLP that enforces them.
What do AI consulting services for government agencies include?
Ours cover four steps: a planning meeting, an AI use policy and plan aligned to the NIST AI RMF, the first build inside your Microsoft 365, and running it afterwards with monthly reporting. Government AI solutions are priced from $25K to $60K for a first engagement.
Questions buyers ask
Before you decide
Will our data be used to train AI models?
No. Your data is never used to train AI models. We deploy OpenAI, Anthropic and open-source models in Microsoft Foundry on Azure. Microsoft’s terms exclude your prompts and outputs from training, and Anthropic’s commercial terms do the same for its Claude models, which Anthropic operates on Azure infrastructure.
We are on GCC or GCC High. Can we still use AI?
Yes. We are an AOS-G partner. AI services reach government clouds at different times, so the assessment confirms what is available in your cloud today and plans around anything that is not.
Do we have to use Microsoft Copilot?
No. Copilot is one option. We pick the model and the tool for each use, including custom agents in Microsoft Foundry and Copilot Studio.
Find out what AI would find in your files, before anyone turns it on
Start AI for local government with a free, one-week, read-only assessment: an access map, a shortlist of uses and a fixed-price plan.


