Industry Guide: Government
How to Use Claude in Government: A Guide for Public Sector Teams
Faster drafts, the same records and classification rules.
This guide is for policy staff, constituent-services and casework teams, procurement officers, records officers, and public-sector IT teams who want to put Anthropic's Claude to work on the memos, correspondence, and procurement documents that move government work forward. It covers Anthropic's own government releases and what Anthropic's solutions and news pages say about agencies using Claude.
Knowing how to use Claude in government means focusing it on document-heavy, record-bound work: summarizing proposed rules and policy changes in plain language, drafting constituent correspondence and casework notes from an intake file, reviewing RFP responses and procurement documents against stated criteria, triaging public-records and FOIA requests, drafting meeting minutes and public notices, and reviewing legacy systems for security gaps. Claude drafts, summarizes, and flags inconsistencies. Caseworkers, policy staff, records officers, and procurement officials keep every eligibility, legal, and disclosure decision, and agencies run AI-assisted work inside the data classification and records-retention rules they already follow.
Published
Free lesson · no account neededThe Iteration LoopMaster the Essentials6 minStart the free lessonKey Takeaways
- The strongest public-sector uses are policy and regulation summarization, constituent correspondence and casework, procurement document review, records and FOIA triage, and meeting-minute drafting.
- Anthropic's Claude for Government runs inside a separate, FedRAMP High authorized boundary built on Palantir's FedStart platform, for federal, state, and local agencies.
- Anthropic's September 2026 general-availability announcement names memo creation, RFP reviews, and casework as desktop workflows agencies use today.
- Anthropic's case study on the Government of Alberta describes a ministry team using Claude Code to review hundreds of millions of lines of government code for security gaps, with engineers approving every patch.
- Eligibility decisions, legal interpretation, and what gets filed, published, or disclosed under a public-records request stay with the staff member or official responsible for that decision.
Copy and Paste
Prompts You Can Use Today
Plain-language rewrite of a proposed rule
Rewrite the regulatory text pasted below in plain language for a public-facing summary, at roughly an eighth-grade reading level. Preserve every substantive requirement and deadline exactly, flag any sentence where simplifying risks losing a legal nuance, and list the defined terms from the original text that the plain-language version still needs to explain. Regulatory text: [paste].
Flagging nuance risk is what lets a policy writer trust the simplification instead of rereading the original line by line.
Constituent correspondence from a case file
Draft a response letter to the constituent using only the case notes and policy reference pasted below. State the current status of their request exactly as the notes describe it, cite the policy section for any eligibility or timeline statement, use a respectful and plain tone, and flag any detail the case notes are missing before this letter can go out. Case notes and policy reference: [paste].
Grounding every status claim in the actual case notes keeps the letter accurate without the caseworker rereading the full file.
Public-records request triage
Review the records request pasted below and extract what records are being requested, the date range, the requesting party, and the statutory response deadline that applies. Flag any part of the request that looks overly broad or that references a record type our office does not hold, and list what our intake checklist requires that is missing from the request as written. Records request: [paste].
This speeds up intake without letting the model decide whether an exemption actually applies to any record.
RFP response review against stated criteria
Compare the vendor proposal pasted below against the evaluation criteria and minimum requirements from our RFP, also pasted below. List each criterion, whether the proposal meets it, and the page or section of the proposal that supports your answer. Flag any required element the proposal does not address at all. RFP criteria and vendor proposal: [paste].
Citing the exact page for every match is what lets a procurement officer verify the comparison quickly instead of rereading the full proposal.
Meeting minutes from a recording transcript
Draft meeting minutes from the transcript pasted below: list attendees, each agenda item discussed, the decision or action reached on each item, who is responsible for the follow-up, and the stated deadline. Do not characterize disagreement or attribute a stance to anyone not directly quoted in the transcript. Transcript: [paste].
Sticking to only what the transcript directly supports keeps the minutes safe to circulate without a chair's edit first.
From Anthropic
What Anthropic offers for Government
Claude for Government
Anthropic's solutions page for government describes Claude for Government as running inside a separate, FedRAMP High authorized boundary built on Palantir's FedStart platform, available to U.S. federal, state, and local government agencies and qualifying public-sector organizations. The page describes document review, research, drafting, and case processing as core uses, admin controls including an admin console, identity-provider sign-in, spend caps, and exportable audit records, prepaid usage pricing with no per-seat fee, and procurement through Carahsoft via the GSA Multiple Award Schedule, NASPO ValuePoint, and OMNIA Partners, plus a California statewide contract.
Claude for Government is now generally available
Anthropic's September 2026 article announces that Claude for Government moved out of a public beta that started in July 2026 to general availability for federal and state agencies, delivering Claude's capabilities in a FedRAMP High authorized environment. It names memo creation, RFP reviews, and casework as desktop workflows, describes Claude Code as used for modernizing the software systems behind public services, and notes Claude Code CLI and Claude for Microsoft 365 remain in early access.
· Source
In Practice
How Government Teams Use Claude
According to Anthropic's published customer story
Government of Alberta
Anthropic's case study describes an internal team inside Alberta's Ministry of Technology and Innovation, set up in 2025, using Claude Code with Claude Opus and Sonnet models to review government systems for security vulnerabilities. The team scanned about 466 million lines of code across roughly 1,280 applications and 3,400 repositories in about 20 hours using around 50 autonomous agents, against an estimated 6.5 years for the same review done manually, running Claude-based review agents built on the Claude Agent SDK that checked each application against roughly 95 security controls per pass. Anthropic reports the team rebuilt a subsidy portal originally hand-coded in Java about 25 years ago, which took five months to build the first time, in as little as four to five days in some scenarios, and that one ministry plans to consolidate 185 legacy applications into 16 reusable ones.
Illustrative walkthrough
A records request moves from intake to response
- A records officer pastes an incoming public-records request into a shared workspace along with the office's intake checklist and the statute setting the response deadline.
- Claude extracts the records being requested, the date range, and the deadline, and flags that the request references a record type the office does not hold.
- The records officer sends a clarifying question to the requester and confirms which records actually exist before any search begins.
- Once the records are located, Claude drafts a response letter citing which records are enclosed and which portions are withheld under a specific exemption the officer has already identified.
- The records officer reviews the letter against the file, confirms the exemption citation, and signs the final response before it goes out.
Core Workflows
Claude in Government Use Cases From Intake to Disclosure
These are the areas where AI for government teams saves the most staff time while keeping records, classification, and accountability rules intact.
Policy and regulation summarization
AI for government policy work turns a dense rule change or bill into a plain-language summary that still names every deadline and requirement, so staff can brief leadership faster.
Constituent correspondence and casework
Drafting responses to residents from the actual case notes and policy reference, so a caseworker reviews a grounded draft instead of starting every letter from a blank page.
RFP and procurement review
Checking a vendor proposal against the stated evaluation criteria line by line, with a citation to the supporting page, ahead of a procurement officer's own review.
Public-records and FOIA triage
Extracting what a records request actually asks for, the applicable deadline, and what the intake checklist still needs, without deciding whether an exemption applies.
Meeting minutes and public notices
Turning a recording or transcript into minutes or a public notice that sticks to what was actually said and decided, ready for a chair's review.
Legacy system and security review
Anthropic's case study on the Government of Alberta describes using Claude Code to scan legacy government code for vulnerabilities, with engineers approving every proposed patch before it ships.
What Agencies Report
How Public-Sector Teams Use AI Today
How is AI used in government? Anthropic's September 2026 general-availability announcement for Claude for Government names memo creation, RFP reviews, and casework as workflows agencies already run through the Claude desktop app, alongside Claude Code for modernizing the software systems that underpin public services.
The clearest documented example is Anthropic's case study on the Government of Alberta, where an internal team inside the Ministry of Technology and Innovation used Claude Code, running Claude Opus and Sonnet models, to scan roughly 1,280 applications across about 3,400 code repositories, around 466 million lines of code in about 20 hours using roughly 50 autonomous agents, against an estimated 6.5 years for the same review done manually. The team ran Claude-based review agents checking each application against roughly 95 security controls per pass, and reports rebuilding a 25-year-old subsidy portal, originally built in about five months, in as little as four to five days in some scenarios, with one ministry planning to consolidate 185 legacy applications into 16.
Across AI for government agencies, AI for state and local government, and AI in the public sector more broadly, the pattern Anthropic describes is document-heavy, record-bound work: summarizing, drafting, and reviewing against an existing standard, while the people accountable for an eligibility decision, a legal position, or a records disclosure keep making that call themselves. The generative AI in government conversation tends to focus on exactly these tasks rather than on fully automated decisions.
Anthropic's Releases
The Government Tools Anthropic Has Built
Anthropic's solutions page for government describes Claude for Government as running inside a separate, FedRAMP High authorized boundary built on Palantir's FedStart platform, available to U.S. federal, state, and local government agencies and qualifying public-sector organizations. The page describes document review, research, drafting, and case processing as core uses, alongside admin controls such as an admin console, identity-provider sign-in, spend caps by user tier, and exportable audit records to support authorization-to-operate and inspector-general documentation.
Anthropic's September 2026 article announcing general availability says the platform moved out of a public beta that started in July 2026 and is now generally available for federal and state agencies, with new features arriving on the same release cadence as Anthropic's commercial product. The article describes Claude Code CLI and Claude for Microsoft 365 as still in early access rather than general availability, and notes agencies pay for prepaid usage in fixed increments with no per-seat fee and no seat minimum.
Anthropic's solutions page also describes procurement paths through Carahsoft, including the GSA Multiple Award Schedule, NASPO ValuePoint, and OMNIA Partners, plus a California statewide contract administered by the state's Department of Technology that is available to state entities and local governments. Among government AI tools sold through these vehicles, Claude for Government is one Anthropic can point an agency's procurement office to directly.
Writing in Practice
AI for Constituent Correspondence Done Well
The pattern that holds up for correspondence and casework is a named source document and hard references. Give Claude the case notes, the policy section that governs the request, and ask it to cite the specific section for every status or eligibility statement rather than summarize from memory. A draft built that way can be checked against the source in minutes.
Plain-language rewriting of a rule or notice works the same way: preserve every deadline and requirement exactly, and ask Claude to flag any sentence where simplifying the language risks losing a legal distinction the original text relied on. A policy writer reviewing a flagged list moves faster than someone rereading the entire original text line by line.
Keep the division of labor explicit. Claude drafts, summarizes, and extracts. The caseworker, policy analyst, or records officer owns the accuracy of the draft and any judgment call the source material does not fully resolve, and the official of record owns what ultimately gets sent, published, or disclosed.
Choosing Tools
Claude vs ChatGPT for Government Work
The Claude vs ChatGPT for government comparison is most useful run on your own documents: a case file already resolved, a records request already answered, a procurement comparison already scored by a human reviewer. Compare accuracy against the known-correct outcome, how each tool handles a gap in the source material, and whether it invents a detail instead of flagging the gap.
In public-sector practice the decision usually comes down to the authorization boundary and procurement vehicle rather than a single benchmark: whether a deployment is authorized for the data classification your agency handles, which contract vehicle your procurement office already has in place, and whether the tool connects to the case-management, records, or document systems your staff already use. That governance question matters more for AI for the public sector than for a consumer use case, because the data involved is rarely your own.
Is Claude authorized for sensitive government data? Anthropic's solutions page describes Claude for Government as FedRAMP High authorized and says it is built for controlled unclassified information and FIPS 199 High Impact data, with ITAR-controlled data routed instead through Amazon Bedrock in AWS GovCloud. Whether a specific deployment meets your agency's own authorization-to-operate and data-classification requirements is a question for your agency's security and compliance office to confirm, not a general yes.
Getting Started
A Safe First Month for a Government Team
Start with closed cases
Test prompts on a case or request that is already resolved, and compare the AI-assisted draft against what was actually sent or published.
Load your policy references
Put the regulation, program guide, or RFP criteria you draft against into the workspace so drafts and reviews cite the version your office actually uses.
Pick one document type
Choose one high-volume task, such as correspondence drafting or RFP scoring, and measure the change in review time before expanding further.
Confirm the authorization boundary
Agree with your security and compliance office which data classification can touch which deployment before any case file or records request goes near it.
Limits
Where Public Officials Must Decide
Claude can misstate a policy detail, miss a deadline buried in a dense regulation, or drift into a legal conclusion the source material does not support. Every AI-assisted letter, summary, or procurement comparison needs a reviewer who checks it against the actual case file, regulation, or proposal before it goes anywhere near a constituent, a vendor, or a decision-maker.
Does using AI put sensitive government data at risk? The risk is not the drafting assistance itself; it is where case files, records requests, or classified and controlled information are sent and processed, and whether that deployment is one your agency's security office has actually authorized for that data's classification. Eligibility determinations, legal interpretation, and what ultimately gets filed, published, or disclosed stay with the accountable official.
Treat case files, records requests, and procurement evaluations as sensitive by default. Use only an authorized deployment for the data classification involved, follow your agency's retention and disclosure rules, and keep the final record of what was sent, decided, or published inside your normal case-management and records systems, not inside a chat transcript.
Build Reliable AI Habits for Public Sector Work
Getting trustworthy drafts from AI is a practiced skill: grounding, citation discipline, and review. The free Nightschool AI curriculum is hands-on from the first lesson.
Frequently Asked Questions
How is AI used in government?
Mostly for document-heavy, record-bound work: summarizing policy and regulatory changes, drafting constituent correspondence and casework notes, reviewing procurement proposals against stated criteria, triaging records requests, and drafting meeting minutes. Anthropic's general-availability announcement names memo creation, RFP reviews, and casework as desktop workflows agencies already run. Staff keep reviewing the output and own every eligibility, legal, and disclosure decision.
Can AI help with government records and FOIA requests?
Yes, for extracting what a request actually asks for, the applicable date range, and the statutory response deadline, and for flagging a request that looks overly broad. A records officer still decides whether any exemption applies and what ultimately gets released; the model only speeds up intake and triage.
Is Claude authorized for government use?
Anthropic's solutions page describes Claude for Government as running inside a separate, FedRAMP High authorized boundary built on Palantir's FedStart platform, available to U.S. federal, state, and local agencies. Whether a specific deployment meets your own agency's authorization-to-operate requirements is a question for your security and compliance office, not a general yes.
How much time can AI save for government agencies?
It depends heavily on the document type and how clear the source material is, so measure it on your own cases. Anthropic's case study on the Government of Alberta reports a code security review that took about 20 hours against an estimated 6.5 years done manually, but that figure is specific to a large-scale code-scanning project, not a general correspondence or casework estimate.
What are the biggest AI use cases in the public sector?
The most common are policy and regulation summarization, constituent correspondence and casework, procurement document review, records and FOIA triage, and meeting-minute drafting. Each works best with a named source document, such as a case file or an RFP, and a staff reviewer checking the output before it is relied on.
Can AI help modernize legacy government systems?
Anthropic's case study on the Government of Alberta describes a ministry team using Claude Code to scan hundreds of millions of lines of legacy code for security gaps and propose fixes, with engineers approving every patch before it ships. That is a software-modernization use case distinct from the policy and casework workflows this guide focuses on.
Does using AI put sensitive government data at risk?
The risk comes from where case files, records, or classified and controlled information are sent and processed, not from drafting assistance itself. Use only a deployment your agency's security office has authorized for that data's classification, and keep eligibility, legal, and disclosure decisions with the accountable official.
How does Claude compare to ChatGPT for government work?
Run the comparison on your own finished work: a resolved case file, an answered records request, a procurement comparison already scored by a human reviewer. Compare accuracy, how each tool handles a gap in the source material, and whether it invents a detail. In practice, the deciding factors are usually the authorization boundary and procurement vehicle, not a single benchmark.
Keep going
Related Guides and Courses
Related reading
- AI Training for Work
- Free AI Prompt Generator for Work Tasks
- How to Use Claude in Real Estate
- How to Use Claude in Education
- How to Use Claude in Marketing
- How to Use Claude in Manufacturing
- How to Use Claude in Retail and Ecommerce
- How to Use Claude in Logistics
- Private Equity Due Diligence With Claude: Data Rooms to Findings
- How to Do CIM Analysis With Claude: Screening Deals and Drafting the IC Memo
- LBO Modeling With Claude: What It Builds, What You Still Have to Check
- Venture Capital Deal Sourcing With Claude: A Thesis-Driven Workflow