Idea
Governments use AI to process cases, detect fraud, answer questions, translate, draft documents and support decisions — sometimes taking decisions outright.
Theory
Advocates expect cheaper and faster services and better targeting. Public-law scholars stress that automated decisions must remain explainable, contestable and attributable to an accountable authority.
Real examples
- The Dutch childcare-benefits scandal involved risk-based fraud detection and led to the cabinet's resignation in January 2021.
- Australia's Robodebt scheme, which raised welfare debts by income averaging, was the subject of a Royal Commission that reported in 2023.
- The EU AI Act (2024) treats many public-sector uses as high-risk.
Critique
Automation bias, opaque models, poor data and weak appeal routes can turn errors into mass harm. Efficiency gains are often asserted rather than evaluated.
Results
The OECD Digital Government Outlook 2026 reports near-universal AI use among surveyed governments but far less systematic evaluation. Binding rules on public-sector AI are now emerging in the EU and the Council of Europe.
Related
Real examples
An AI system launched in January 2025 as the e-Albania virtual assistant. In September 2025 the government presented it as a 'virtual minister' for public procurement; its legal status is disputed and the Constitutional Court did not rule on its merits.
A Chilean government project (1971-73) led by Fernando Flores with the British cybernetician Stafford Beer to link state-run firms to Santiago by telex, statistical software and an operations room. Prototypes and pilots were built; the 1973 coup ended it.
A nationwide computer network proposed from 1962 by Soviet cyberneticist Viktor Glushkov to gather data and help plan and manage the USSR's economy. The Politburo reviewed it on 1 October 1970; it was never funded in full or built.
Claims
AI was used in at least one government function in 35 of the 36 OECD countries surveyed, but only 10 of the 36 reported any evaluation of these uses.[1]
In the Dutch childcare-benefits scandal, tax authorities wrongly treated thousands of families as fraudsters, partly through risk-scoring practices; the cabinet resigned in January 2021.[1]
The UK government mandated its Algorithmic Transparency Recording Standard across government departments on 6 February 2024, and on 17 December 2024 published a policy defining which bodies and algorithmic tools must publish records.[1]
Revision history
- 2026-10-23rev 2 — Seed import
- 2026-10-23rev 1 — Seed import