⚡ Quick Summary
Published by the OECD as No. 20 of its Artificial Intelligence Papers series, this paper was written by Barbara-Chiara Ubaldi and Ricardo Zapata under the supervision of Carlos Santiso, and approved and declassified for publication on 11 June 2024.
It examines how governments govern with AI, arguing that beyond their regulatory role they are also users and, in some cases, developers of AI systems, a role that has received less attention. It sets out three expected impacts — productivity, responsiveness and accountability — and analyses 71 use cases from 31 Member and accession countries collected through the OECD Survey on Digital Government 2.0 and the OPSI Call for Innovations in Government. Table 1 pairs AI system tasks with four government functions: internal operations, policy making, service delivery, and internal and external oversight.
Section 3 reviews policy challenges and country responses — strategic objectives, policy design, implementation support, and monitoring and oversight — and the paper ends with a preliminary framework built on three policy questions (what, who, why) and four policy measures: enablers, guardrails, engagement and impact.
🧩 What’s Covered
The paper moves from evidence on current use to national policy responses and a proposed framework.
- Section 1 – Benefits and risks: productivity (Nijmegen's city-centre counting, Queensland's land-use mapping), responsiveness (Norway's Frida chatbot, resolving 80% of enquiries) and accountability (Transport Canada's cargo risk assessment), then risks including bias amplification, opaque design, and privacy and security breaches, with attention to law enforcement, immigration control, welfare benefits and fraud prevention.
- Section 2 – Delivering impact: Table 1 pairs AI tasks with four functions (efficiency, effectiveness, responsiveness, accountability) and with impacts; the 2023 OECD Digital Government Index found about 70% of countries using AI for internal operations, 30% for policymaking and 67% for service design and delivery.
- Box 2.1 – Tasks of AI systems: definitions and examples of the taxonomy's tasks, from recognition, event detection, forecasting and personalisation to interaction support, goal-driven optimisation, reasoning with knowledge structures and content generation.
- Country examples by function: France's "Albert", Sweden's email-sorting model, Italy's GiusBERTo, Korea's infectious-disease forecasting system, Finland's AuroraAI, Austria's Mona chatbot, Türkiye's "Neyim Var?", Spain's grant-fraud detection and Brazil's Alice procurement analyser.
- Section 3 – Policy issues and actions: strategic objectives; institutional arrangements (European AI Office, Spain's AI Supervision Agency, US Chief AI Officers); soft guardrails (Australia's AI Ethics Framework, Canada's Guide on generative AI, Japan's AI Guidelines for Business); regulation (Canada's Directive on Automated Decision-Making, the 2024 US Executive Order); procurement (IEEE P3119, EU model contractual AI clauses); and oversight (algorithm registers, Norway's Office of the Auditor General, Chile's Transparency Council).
- Box 2 – EU AI Act: the Act's four risk levels (unacceptable, high, limited, minimal), obligations for high-risk systems such as risk management and fundamental rights impact assessment, and the governance framework of national competent authorities, the European Artificial Intelligence Board, an advisory forum and the European AI Office.
- Sections 4 and 5 – Framework and conclusions: the preliminary framework in Table 2 and Figure 1, organised around the three policy questions and the measures of enablers, guardrails, engagement and impact, and conclusions calling for more consistent approaches, better evidence and indicators, and a possible global repository.
💡 Why it matters?
For public sector leaders and oversight bodies, the paper supplies a shared vocabulary and an evidence base for deciding where AI is worth deploying and what must accompany it. Its taxonomy separates what an AI system does from the government function it serves, which helps assess a proposed use against productivity, responsiveness and accountability goals rather than treating AI as one thing. The catalogue of institutional, regulatory, procurement and monitoring measures shows how peers have addressed similar constraints, and the framework offers a structure for organising enablers, guardrails and engagement. It links its analysis to international instruments the document itself cites, including the OECD AI Principles and the EU AI Act.
❓ What’s Missing
Much of the evidence rests on survey data and a pool of 71 use cases that is not reproduced in the document, so individual cases cannot be inspected. Impact measurement is described as an area of opportunity: the paper states that there is no robust track record of systematic use or of the human and financial resources needed to scale initial efforts. The framework is explicitly preliminary. The mandate of new supervisory institutions in relation to public sector AI is noted as not always clearly defined, and enforcement is described as a challenge, with a minority of strategies setting implementation timeframes.
👥 Best For
Best for civil servants and digital government units designing or procuring AI, for audit and oversight bodies assessing public sector use, and for policy teams preparing AI strategies where government is the deployer rather than only the regulated subject. Researchers and advisers comparing national approaches will find the country examples and framework useful.
📄 Source Details
Governing with Artificial Intelligence: Are governments ready?, OECD Artificial Intelligence Papers No. 20, published by the OECD, June 2024. Written by Barbara-Chiara Ubaldi and Ricardo Zapata under the supervision of Carlos Santiso; approved and declassified by the OECD Public Governance Committee on 11 June 2024. The text runs to 31 numbered pages (32 PDF pages) and includes references and endnotes. Language: English. No URL for the document itself is printed in the text; the extraction covered all 32 pages.
