Artificial intelligence in government is the use of AI by public bodies to answer citizen inquiries, process claims, detect fraud and plan services. Federal agencies, state and local government and national administrations already run AI tools for these jobs. Rules such as the EU AI Act and federal policy require risk checks and human oversight.
What you’ll take from this
This guide is for public servants, government leaders and suppliers who want a clear picture before they implement AI. If you will be accountable for those decisions, the Responsible AI Leadership online program from Founderz covers risk assessment, transparency and adoption in practice.
- Know where the public sector uses AI today and where it still struggles.
- Compare how the U.S., the EU and the UK govern AI in public bodies.
- Learn from real programs in Finland, Estonia and the UK, and from one costly failure.
- Follow a six-step approach to AI integration.
What is artificial intelligence in government?
It is the use of AI technology, such as machine learning, natural language processing and predictive analytics, to run government processes and deliver public services.
The scope reaches all levels of government. The OECD report Governing with Artificial Intelligence (2025) reviews 200 real-world examples across 11 core government functions, from public finances to justice and anti-corruption work. Many AI solutions arrive inside software an agency already buys. That is why the rules focus on use.
How do government agencies use AI to streamline service delivery?
Government organizations use AI to automate routine work, answer government inquiries faster, detect fraud and support decisions with data. Here are the main uses and the risk each one carries.
| Use | What it does | Risk to manage |
|---|---|---|
| Citizen services chatbot | Answers questions on benefits, taxes or permits at any hour | Wrong answers given with confidence |
| Case processing | Reads forms, extracts data and routes cases to the right team | One error copied across thousands of files |
| Fraud detection | Flags unusual claims or payments for human review | False alarms that hit honest people |
| Forecasting | Predicts demand for services or emergency response times | Biased historical data shaping future decisions |
| Drafting and summaries | Helps write letters, briefings and meeting notes | Sensitive records shared with the wrong tool |
| Computer security | Spots attacks on networks and critical infrastructure | Attackers targeting the AI system itself |
More than 20,000 UK civil servants took part in a government-led trial of AI-powered tools. Microsoft 365 Copilot users saved an average of 26 minutes a day on self-reported figures, according to the UK government’s June 2025 announcement.
What is the government version of ChatGPT?
There is no single government ChatGPT. Public bodies use the same large language models as the private sector, in versions with stricter security and data controls, or build their own assistants on them.
In January 2025 OpenAI launched ChatGPT Gov, which American agencies at any level can deploy in their own Microsoft Azure environment. OpenAI said that since 2024, more than 90,000 users at over 3,500 agencies had sent over 18 million messages on ChatGPT. Estonia took another route. Its Bürokratt program links public authorities’ chatbots into one network covering around 3,000 e-services, as the European Commission’s Open Source Observatory describes it.
Examples of introducing AI across government sectors
Finland ran AuroraAI from February 2020 to the end of 2022. The national AI program, led by the Ministry of Finance, organized services around life events, such as changing jobs, instead of around agency structures.
Budgets show how much governments expect from the potential of AI. Public spending on AI in the United States passed $20 billion in 2024, mostly through grants and contracts. European public AI contracts reached $3.66 billion, led by the UK, according to the Stanford AI Index 2026.
The warning comes from Australia. Robodebt (2016 to 2020) used automation and data-matching to issue welfare debt notices, many of them wrong. A court-approved settlement in 2021 was worth about A$1.8 billion, and a Royal Commission called it a “costly failure of public administration”, as the Wikipedia entry on Robodebt records. Software decided about people with no real human check.
What rules govern AI in the public sector?
Public bodies face stricter rules than most companies because their decisions affect rights, benefits and safety. The main frameworks in America, the EU and the UK all demand risk checks, transparency and human oversight.
| Jurisdiction | Main instrument | What it asks of government |
|---|---|---|
| United States (federal) | OMB memo M-25-21 (April 2025) | A Chief AI Officer per agency, annual use case inventories, and testing and human oversight for high-impact AI |
| European Union | AI Act, Regulation (EU) 2024/1689, amended by Regulation (EU) 2026/1744 | High-risk rules for uses such as eligibility for benefits, plus a fundamental rights impact assessment, from 2 December 2027 |
| United Kingdom | Algorithmic Transparency Recording Standard | Published records for in-scope algorithmic tools, mandatory for central government since 2024 |
The federal rules sit in OMB memorandum M-25-21, which the Office of Management and Budget issued in April 2025 to replace M-24-10. Congress had set the direction with the AI in Government Act of 2020. The memo treats AI as high-impact when its output is the main basis for decisions with a significant effect on rights or access to services.
OMB memos do not bind state or city governments, which write their own policies. Any of them can use the voluntary NIST AI Risk Management Framework, released in January 2023.
In Europe, Annex III of the AI Act lists systems used by public authorities to evaluate eligibility for public assistance benefits and services as high-risk. Article 27 requires public bodies, and private entities providing public services, to carry out a fundamental rights impact assessment before deployment. The Digital Omnibus on AI, in force since 27 July 2026, moved these obligations to 2 December 2027. Social scoring has been banned since February 2025.
What are the risks of AI in government?
The main risks are biased or wrong decisions at scale, data privacy failures, weak information security and lost public trust when people cannot challenge a decision. Budget constraints make each harder to manage.
- Test outcomes by group before launch, because a data set built from past decisions repeats past unfairness.
- Limit what an AI tool can see in citizen records and log every access.
- Treat each AI system as a security target, with reliable data sources and tracked data provenance.
- Name an owner for every AI system who can explain its output to the person affected.
In May 2025 the Cybersecurity and Infrastructure Security Agency (CISA), part of the Department of Homeland Security, published AI data security guidance with the NSA and the FBI. Our article on cyber intelligence covers the defensive side. The Government Accountability Office (GAO) groups the wider checks in its 2021 accountability framework under governance, data, performance and monitoring.
How can state government and federal agencies integrate AI responsibly?
Start with one well-defined problem, set governance before you buy, test with real data and keep a person accountable for each decision the AI supports. These best practices fit any AI project.
- Pick a process with a clear cost, such as a permit backlog.
- Establish governance first: name an owner, a risk lead and the rules for personal data and security.
- Write procurement requirements that ask vendors for testing results, data handling terms and the right to audit.
- Run a pilot with the staff who do the work, then compare time saved and error rates with the manual process.
- Publish what the system does and how people can ask for a human review.
- Monitor performance after launch, since data drift and new fraud patterns change how AI models behave.
Here is a worked example. A city that gets thousands of emailed questions a month sets up an AI assistant for the 20 most common ones, such as waste collection days, and routes the rest to staff. The pilot expands only when inquiries closed without a hand-off rise and complaints stay low.
The workforce decides how far government efficiency gains go, and training existing employees in AI basics and risk is faster to arrange than hiring scarce AI talent. For task-by-task ideas, see our guide to AI use cases for government officials.
Further reading
Sources on how government entities use AI.
- OECD report on governing with AI (2025)
- AI Risk Management Framework (NIST)
- Regulation of artificial intelligence (Wikipedia)
Frequently asked questions
Short answers to the questions public servants ask most often about AI in government.
Can AI be controlled by the government?
Yes, through law, procurement and oversight. The EU AI Act sets binding rules for AI systems placed on the EU market. Federal policy governs how American agencies use AI, and any government can attach conditions to its contracts with AI vendors.
How can AI be used in governance?
AI can analyze responses to public consultations, forecast the effects of a policy, spot fraud in public spending, summarize legislation and plan government operations. It should inform decisions made by accountable people, who can explain and defend them to the public.
What does AI in state and local government look like?
Common uses include website chatbots, translation of public information, document processing and checks on benefit claims. These bodies set their own AI policies because the OMB memos written for federal departments do not apply to them.
Government AI needs buyers who know where to check it. The Responsible AI Leadership online program at Founderz teaches professionals to assess AI risks, ensure transparency and lead responsible AI adoption. It runs online over two months.
