The AI Agent Government Tracker
A verified list of governments shipping AI into public services. Primary sources only. Updated as governments ship.
Why this tracker exists
On February 25, 2026, France's open-data agency shipped something quietly historic. They published an MCP (Model Context Protocol) server at mcp.data.gouv.fr/mcp that lets Claude, ChatGPT, and Gemini query French public datasets through natural conversation, directly, without integrations.
The server crashed in its first days. Too much traffic. The French redirect page read sorry.data.gouv.fr.
That moment went viral for a reason. A government had just proven something every SMB founder now has to reckon with: your data, your workflows, your tools are either reachable by AI agents or they are not. The middle ground is shrinking.
We went looking for every government that has crossed this line. Three tiers, in descending order of technical advancement, are tracked below.
Government-shipped MCP servers
Agent-addressable APIs. An AI model can discover and call these with no custom glue code. Across roughly fifty serious candidates, only three government bodies have shipped one into public preview or production.
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
France - DINUM (data.gouv.fr) data.gouv.fr MCP server | MCP server exposing the French open-data catalog to AI clients at mcp.data.gouv.fr/mcp. | Feb 25, 2026 | |
United States - Government Publishing Office GovInfo MCP server | MCP server for GovInfo data: congressional documents, Federal Register, Supreme Court opinions. | Jan 22, 2026 (public preview) | |
United States - Census Bureau Census Data API MCP | Official MCP server for the Census Data API: ACS, decennial census, economic indicators. | Mar 6, 2026 (beta) |
Citizen-facing AI agents in production
Chatbots, assistants, and voice systems serving the public directly. Quality varies wildly. Below are the deployments with at least one verifiable primary source.
Europe
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
United Kingdom - Government Digital Service GOV.UK Chat | Claude-powered assistant via AWS Bedrock; 10K-user pilot, now rolling into the GOV.UK app. | Dec 2024 pilot, expanded Mar 2026 | |
Estonia - RIA Bürokratt | Network of virtual assistants integrated across 18 public bodies, moving to per-agency agents in 2026. | Production since 2022 | |
France - DINUM Albert | Sovereign LLM assistant for France Services advisors; deployed in France's civil courts. | Apr 2024 | |
Finland - Kela (Social Insurance Institution) Kela-Kelpo and FPA-Folke | Benefits chatbots; Nordic region's largest conversational-AI migration (IBM Watson to boost.ai). | Mar 2021 | |
Norway - NAV Frida | Chatbot for pensions, unemployment, sickness benefits; handled 270K COVID-era inquiries at ~80% resolution. | Summer 2018 | |
Sweden - Skatteverket Skatti | Chatbot for personal tax and population-registration questions. | 2021 | |
Denmark - Digitaliseringsstyrelsen Børge | Generative-AI editor assistant rewriting borger.dk content to plain-language standards across 1,200 pages, built on Claude 3.5 Sonnet. | Feb 2025 | |
Iceland - Government of Iceland Icelandic GPT-4 partnership | Partnership with OpenAI: GPT-4 trained on Icelandic via RLHF with 40 volunteer trainers, making Icelandic the first non-English language in GPT-4's RLHF program. | Mar 2023 | |
Netherlands - SSC-ICT Vlam-chat | Sovereign AI assistant for eight ministries and ~26,500 civil servants. | 2026 |
Middle East
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
United Arab Emirates - Federal Government U-Ask | Unified Arabic/English generative-AI chatbot for UAE federal services. | 2023 | |
United Arab Emirates - Digital Dubai Rashid | AI smart-city advisor for Dubai services. | Production |
Asia-Pacific
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
Singapore - GovTech VICA | Platform of 100+ chatbots across 60+ agencies, serving 800K+ monthly queries. | Production | |
Singapore - Open Government Products Pair | LLM assistant for public officers (drafting, research), plus the LaunchPad innovation platform. | Production | |
Japan - Digital Agency Gennai | OpenAI-powered internal assistant; ~180K personnel target. | Rolling 2025-2026 | |
Japan - Tokyo Metropolitan Government Bureau-wide ChatGPT deployment | ChatGPT deployed to ~50K bureau employees, now expanded to all bureaus. | Aug 2023, expanded Mar 2026 | |
South Korea - Seoul Metropolitan Government Seoul Talk 2.0 | Citizen chatbot (500K+ users) being upgraded to generative AI. | Chatbot 2.0 rolling 2025 | |
India - Central Government + AI4Bharat + Microsoft Jugalbandi | WhatsApp generative-AI assistant covering 171 government schemes across 10 Indian languages. | Rolling since Apr 2023 | |
Taiwan - MODA TAIDE + Government AI Application Sandbox | Sovereign chatbot and a testing sandbox running 10+ LLMs for government use cases. | 2024-2025 | |
Australia - NSW Ombudsman Ombudsman AI assistant | AI assistant for ombudsman inquiries. | Production |
North America
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
United States - California (CDT) Poppy | Generative-AI assistant deployed across 20 state departments. | Pilot 2024-2025 | |
United States - City of Denver Sunny | 311 generative-AI chatbot in 72+ languages across web, text, and WhatsApp. | Feb 2024, expanded Oct 2025 | |
United States - New York City MyCity Business Chatbot | Azure OpenAI bot trained on 2,000+ NYC Business pages. Notable for advising businesses to break the law; later scaled back. | Oct 2023 | |
Canada - Shared Services Canada CANChat | In-house multilingual generative-AI assistant for public servants. | Pilot, rolling |
Latin America
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
Argentina - ARCA (formerly AFIP) Tina | Virtual tax assistant on web, mobile, and WhatsApp for fiscal-key recovery, CUIT/monotributo registration, payments. | Production | |
Chile - SII (Internal Revenue) Sofía | Hybrid rule-based and ML tax assistant; second tax authority in LatAm to ship conversational AI. | Mar 2021 | |
Peru - SUNAT Sofía | Citizen-service assistant; 4.9M non-presencial interactions through Oct 2024. | Pre-2023 | |
Colombia - DIAN DIANA | Cognitive virtual tax assistant with ML intent classification. | Production | |
Mexico - IMSS ChatIMSS / Atención VIHrtual / Pension AI | ChatIMSS for rights validity; WhatsApp HIV-care assistant; generative-AI pension assistant on NVIDIA + LangChain. | Pension assistant 2025 | |
Brazil - São Paulo (Prodesp) Poupinha + WhatsApp 'P' | Assistants for Poupatempo services: ID cards, driver's licenses, pensions, vaccination cards. | 2016 pilot, 2018 production, WhatsApp extension after | |
Brazil - Receita Federal BotCPF + BotRTC | BotCPF for citizen tax-ID queries; BotRTC generative-AI chatbot for the tax reform. | BotCPF Jun 2025, BotRTC Feb 2026 | |
Uruguay - AGESIC ID Uruguay assistant + smoking-cessation chatbot | Virtual assistant for ID Uruguay (digital identity) and a Ministry of Health smoking-cessation chatbot. | Apr 2018 |
Supranational
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
European Union - Publications Office Publio | AI assistant for searching EU law and the EU Whoiswho directory. | Production |
Back-office AI improving public services
Fraud detection, benefits triage, audit targeting. Less visible to citizens but operating at significant scale.
| Country / Agency | What shipped | Launch | Source |
|---|---|---|---|
United Kingdom - HMRC Ask HMRC Online | Digital assistant covering 60+ tax types; 5.48M accesses in 2024; generative-AI trials with 15K-user expansion. | Production 5+ years | |
United Kingdom - DWP Hardship-payment ML models | ML models flagging hardship-payment claims since May 2022; broader fraud detection with documented fairness issues. | Production since 2022 | |
United States - Social Security Administration AI voice / chat triage + HeaRT | Customer-facing AI voice and chat triage; HeaRT hearing-transcription system rolling nationwide. | 2025 | |
France - DGFiP Undeclared-property detection | AI/data-mining for undeclared-property detection, run by a 30-person data-science team since 2014. | Production, expanding | |
Denmark - Udbetaling Danmark + ATP Social-benefits fraud models | Up to ~60 algorithmic models detecting social-benefits fraud; flagged by Amnesty International for discrimination risk. | Production; audit Nov 2024 | |
Sweden - Arbetsförmedlingen NLP labour-market pipeline | NLP pipeline extracting job titles and skills from unstructured ads to power jobseeker matching. | Production | |
Peru - SUNAT Four predictive tax models | Production models: debt recovery (S/3.58B flagged), payment-plan default, shell-company risk, and Atenea audit support. | 2024 | |
Mexico - SAT Fiscal-risk classifier | AI classifying taxpayers by fiscal risk, detecting fake-invoice networks; Q1 2024 audit effectiveness up 98.5% YoY. | 2020 model, 2024 plan | |
Colombia - DIAN Real-time invoicing + collection AI | Real-time electronic-invoicing monitoring across 120K+ establishments; debt-collection prioritization AI identified 8,440 delinquents. | 2024 |
Commonly misreported
These show up in coverage and sound credible until you pull the source. We excluded them. Naming what is not real is part of making the list useful.
What this means if you run a business
1. Your SaaS stack is becoming agent-addressable whether you notice or not.
Seventy percent of major SaaS vendors published a remote MCP server in the last year. Stripe, HubSpot, Notion, Linear, Slack, Intercom, Shopify, Xero. If your accountant, ops lead, or marketer hasn't tried asking Claude to execute a task against these tools yet, you are one conversation away from a capacity multiplier.
2. Early deployments fail publicly, and the sector still ships the next one.
NYC's MyCity told businesses to break the law. It got overhauled and other cities kept shipping. The SSA chatbot has documented glitches. The DWP's fraud AI has documented bias. None of these agencies stopped. The pattern is not "wait until it's safe." The pattern is ship, measure, fix.
3. The working deployments are narrow.
Kela answers benefits questions. Denver's Sunny answers 311 questions. ARCA's Tina recovers fiscal keys. None of these try to be everything. The governments that shipped well picked a bounded domain, locked the scope, and let humans handle the rest. If you're scoping an AI project for your business, do the same.
4. MCP is the wire, and the wire is here.
A year ago "connecting AI to your data" meant hiring an engineer. Today it means installing a server. If you are not thinking about which of your internal workflows should be reachable this way, your competitors either are or will be within the quarter.
How this was compiled
Compiled April 2026 from primary government sources, major press (Reuters, FT, Le Monde, Japan Times, O Globo, Svenska Dagbladet), and verified GitHub repositories owned by government entities. Every entry has at least one primary or major-press source cited. Low-confidence entries are flagged in our internal tracker and omitted from the tables above.
Countries and regions intentionally excluded because evidence was too thin: most of Sub-Saharan Africa, Southeast Asia beyond Singapore, Taiwan, India, most of Central Asia, the Balkans. Deployments may exist. Primary sourcing does not.
Spotted a missing entry, a broken source, or a deployment we got wrong? Email mykhailo.kushnir@destilabs.com.
Thinking about your own MCP-shaped workflow?
DestiLabs builds AI-agent-enabled products and internal tools. Narrow scope, measurable outcome, real infrastructure - the way the governments above shipped for citizens.