Governments are among the largest potential users of AI: they process millions of applications, letters and cases, deliver health and education, and manage cities and infrastructure. AI can make public services faster and more accessible. But public-sector automation has also produced some of the most damaging algorithm-related failures, because errors affect vulnerable people at scale.

Where governments use AI

  • Citizen services: chatbots and assistants answering questions about taxes, benefits and permits, often in multiple languages.
  • Document processing: classifying letters, extracting information from forms and routing cases.
  • Tax administration: detecting errors and fraud, and helping taxpayers file correctly.
  • Health systems: imaging, triage, planning and administration in public hospitals.
  • Urban management: traffic signals, waste collection, infrastructure maintenance.
  • Emergency services: predicting demand and optimizing ambulance deployment.
  • Internal productivity: civil servants using AI assistants to draft, summarize and search.

Lessons from failures

Robodebt (Australia)

From 2016, Australia’s government used an automated system to identify welfare overpayments by averaging annual income data across fortnights — a method that often produced incorrect debts — and placed the burden on recipients to disprove them. The scheme was found unlawful; the government settled a class action in 2020, and a Royal Commission in 2023 described it as a crude and cruel mechanism that caused great harm.

The Dutch childcare benefits scandal

In the Netherlands, tax authorities wrongly accused thousands of families of fraud in childcare benefit claims, demanding repayments that drove many into financial hardship. Investigations found that risk-profiling practices, including the use of nationality as a factor, contributed to discriminatory treatment. The scandal led the Dutch government to resign in January 2021.

Common causes

  • automated decisions without meaningful human review;
  • reversed burden of proof on citizens;
  • discriminatory risk factors;
  • lack of transparency and appeal routes;
  • political pressure to find savings or fraud.

Principles for responsible public AI

  1. Legality: a clear legal basis for automated decisions.
  2. Transparency: public registers of algorithms used and their purposes.
  3. Human oversight, especially for decisions affecting rights and benefits.
  4. Fairness testing before and during deployment.
  5. Contestability: easy ways to challenge decisions.
  6. Accountability: named officials responsible for systems.

The OECD and other international bodies track government AI policies and publish principles for trustworthy AI. In the EU, AI used to assess eligibility for public benefits and services is high-risk under the AI Act.

Generative AI in government

Since 2023, many governments have started giving civil servants generative AI assistants for drafting, summarizing, translating and searching internal knowledge. Several have published guidance on responsible use, typically requiring that:

  • staff verify outputs and remain responsible for content;
  • sensitive or classified information is only used in approved tools;
  • AI is not used to make decisions about individuals without appropriate review;
  • use is recorded where it affects official documents.

Public-facing chatbots raise additional questions: they must give accurate information about rights and obligations, and an incorrect answer about a benefit or a tax deadline can harm people. Good practice includes grounding answers in official guidance, linking to source pages and making it easy to reach a human.

Measuring public value

Governments increasingly evaluate AI projects on outcomes for people — faster processing times, fewer errors, accessibility, satisfaction and fairness — rather than cost savings alone. Independent evaluation and publication of results build public trust.

What to watch next

  • Public algorithm registers in more countries and cities.
  • Implementation of the EU AI Act for public authorities.
  • Generative AI assistants rolled out to civil servants.
  • Court cases challenging automated public decisions.

Key takeaways

  • Governments use AI for citizen services, documents, tax, health and urban management.
  • Robodebt and the Dutch childcare benefits scandal show the harm of flawed automated decisions.
  • Failures shared a lack of human review, reversed burden of proof and poor transparency.
  • Legality, transparency, oversight, fairness and contestability are key principles.
  • AI used for public benefit eligibility is high-risk under the EU AI Act.

Frequently asked questions

Can a government make decisions about me using AI?

Often yes, within legal limits. Many jurisdictions give rights to information, human review or appeal for significant automated decisions.

What was Robodebt?

An Australian automated debt recovery scheme that issued many incorrect welfare debts; it was found unlawful and investigated by a Royal Commission.

What is an algorithm register?

A public list describing algorithms used by government bodies, their purpose and how they work, to increase transparency.