⚡ Quick Summary
Published by OECD Publishing, this report examines how governments are developing and using artificial intelligence to govern, not only how they regulate it. It was approved and declassified by the OECD Public Governance Committee on 05 September 2025 and forms part of the OECD Horizontal Project on Thriving with AI.
The report is built on analysis of dozens of governance approaches and 200 AI use cases across 11 core government functions. It identifies three opportunity areas for government use of AI — productivity, responsiveness and accountability — and groups benefits into four categories: automated, streamlined and tailored processes and services; better decision-making, sense-making and forecasting; enhanced accountability and anomaly detection; and unlocking opportunities for external stakeholders. It then sets out five risk types specific to government use: ethical, operational, exclusion, public resistance and risks of inaction. The central argument is that governments must put in place enablers, guardrails and engagement mechanisms in order to adopt trustworthy AI. Its deliverable is an OECD Framework for Trustworthy Artificial Intelligence in Government, alongside a recommendation that governments prioritise high-benefit, low-risk applications while building maturity.
🧩 What’s Covered
The extracted pages cover the front matter, the executive summary and Chapters 1 and 2; later chapters appear only in the table of contents and in cross-references.
- Foreword and acknowledgements: The report is described as built on analysis of dozens of governance approaches and 200 AI use cases, and as part of the OECD Horizontal Project on Thriving with AI. It cites OECD data from 2023 showing that only 39% of people have moderately high or greater trust in national government, and lists the OECD directorates and named staff who drafted each chapter.
- Executive summary: Key findings are that AI use is most prevalent in public service, justice and civic participation functions and least in policy evaluation, tax administration and civil service reform; that use concentrates on internal operations and service delivery rather than oversight and policymaking; and that classic rules-based and machine learning approaches dominate over generative AI.
- Chapter 1 – How AI is accelerating the digital government journey: Defines an AI system, sets out the three opportunity areas (productivity, responsiveness, accountability) across the policy cycle, and details government-specific risks with examples, including Australia's Robodebt scheme (470 000 incorrect debt notices) and the Netherlands' Toeslagenaffaire (26 000 families wrongly accused). It compares regulatory approaches, including the EU AI Act's four risk levels, the United States' "high-impact" concept in policy M-25-21, and Korea's AI Basic Act taking effect in January 2026.
- Chapter 2 – Trends and early lessons: Reports findings from 200 use cases across 11 functions, with benefit shares (31% improving productivity in analytical tasks, 15% personalisation, 9% automating mundane tasks, 25% anomaly detection, 5% engaging non-governmental actors, 4% unlocking opportunities for external stakeholders) and risk shares (operational 93%, ethical 56%, public resistance 50%, exclusion 38%). It notes that only 61 of 1 343 cases (4.5%) in the European Commission's repository are generative AI.
- Chapters 3 to 5 (table of contents and cross-references only): Chapter 3 covers implementation challenges; Chapter 4 covers enablers, guardrails and engagement, the OECD GovTech Policy Framework and the Framework for Trustworthy AI in Government; Chapter 5 provides deep dives into AI in 11 government functions from tax administration to justice.
💡 Why it matters?
For public servants, auditors and advisers deciding where AI fits in government, the report supplies a shared vocabulary for benefits and risks and a function-by-function view of where adoption is mature and where it is not. Its five risk types make oversight targetable, and it names a risk that is easy to overlook: inaction, which widens the capability gap with the private sector. It also links government deployments to regulatory classifications such as the EU AI Act's risk levels and the United States' high-impact designation, which shape how public sector use cases are assessed.
❓ What’s Missing
Only the first 80 pages of the file were available, so Chapters 3 to 5 are described here from the table of contents rather than read. The report itself concedes that its findings are not generalisable to the wider universe of government AI efforts, that the sample may reflect which initiatives are identified or submitted, and that digital security is excluded from its primary scope. It also notes that public sector research on algorithmic management is scarce. No costed roadmap or measurement method for realised benefits is provided.
👥 Best For
Best for public sector AI leads, oversight bodies and audit institutions mapping where AI is being used across government functions and which risks attach to each; for policy advisers comparing national governance approaches to the EU, US and Korean frameworks; and for analysts designing use case inventories and benefit or risk taxonomies.
📄 Source Details
Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, OECD Publishing, Paris, 2025. ISBN 978-92-64-81828-6 (print), 978-92-64-68405-8 (PDF), 978-92-64-43767-8 (HTML); https://doi.org/10.1787/795de142-en. Language: English. The printed table of contents runs to about page 301; the text available for this review covered pages 1–80 of the 306-page file, comprising front matter, the executive summary and Chapters 1–2.
