Title page 1

Contents 7

Foreword 4

Acknowledgements 5

Executive summary 10

Acronyms and abbreviations 13

1. How artificial intelligence is accelerating the digital government journey 15

The digital government journey 17

Understanding AI's transformative potential 19

Using AI in government, a unique context 22

Key opportunity areas and benefits for AI in government 22

To seize the benefits of AI, its risks need to be managed 32

Realising a positive future for AI in government 44

References 45

Notes 56

2. Trends and early lessons from the use of AI across functions of government 58

OECD analysis of 200 use cases across 11 government functions 59

AI is a priority, but efforts are not systemic 60

General trends in government AI use cases 61

Use cases could pose risks if not implemented in a trustworthy manner 73

References 76

Notes 78

3. Implementation challenges that hinder the strategic use of AI in government 79

Most government AI efforts exist in exploratory or pilot phases, with limited scaling and documentation 80

Common challenges shared across core government functions 83

Challenges that are somewhat less common or vary among government functions 89

References 96

Notes 101

4. Enablers, guardrails and engagement for unlocking trustworthy AI 103

Policy action to unlock AI's potential 104

Strengthening enablers to facilitate the adoption of trustworthy AI 106

Establishing guardrails to guide strategic and responsible AI 131

Engagement to shape strategic and responsible AI 141

A framework for trustworthy AI in government 145

Future OECD work on these issues 146

References 146

Notes 157

5. Deep dive: The current status and future potential of AI in government 164

AI in tax administration 166

Current state of play 166

Managing risks and challenges 170

Untapped potential and way forward 172

AI in public financial management 173

Current state of play 173

Managing risks and challenges 176

Untapped potential and way forward 179

AI in regulatory design and delivery 182

AI in civil service reform 193

AI in public procurement 201

AI in fighting corruption and promoting public integrity 210

AI in policy evaluation 218

Current state of play 218

AI in civic participation and open government 225

AI in public service design and delivery 235

AI in law enforcement and disaster risk management 251

AI in justice administration and access to justice 262

References 275

Notes 298

Tables 9

Table 1.1. Understanding the use of AI in government 23

Table 2.1. Functions of government analysed for Governing with AI 60

Table 2.2. OECD AI Principles 60

Table 2.3. Examples of AI for automated, streamlined and tailored processes and services 68

Table 2.4. Examples of AI for better decision-making, sense-making and forecasting 69

Table 2.5. Examples of AI for enhanced accountability and anomaly detection 70

Table 2.6. Examples of AI unlocking opportunities for external stakeholders through AI as a good for all 71

Table 3.1. Costs for 1 million tokens on common Generative AI models 92

Table 4.1. Policy questions and measures underpinning the Framework for Trustworthy AI in Government 146

Figures 8

Figure 1.1. OECD 2023 Digital Government Index, composite results by country 18

Figure 1.2. Venture capital investments in AI have grown over the years 21

Figure 1.3. AI at each stage of the policy cycle 24

Figure 1.4. More than half (59%) of open-source AI training datasets are in English 41

Figure 1.5. AI incidents have generally trended upwards since late 2022 43

Figure 2.1. Use cases are most present in public service, civic participation and justice functions 62

Figure 2.2. EU and LAC follow a similar trend with the AI use cases sample collected for this report 63

Figure 2.3. AI use cases per core function and government activity 64

Figure 2.4. Public services and engagement represent an important share of use cases under government processes in the EU and LAC 65

Figure 2.5. Potential benefits of AI use cases across functions of government 66

Figure 2.6. Specific benefits of AI use cases 67

Figure 2.7. The potential for operational risks is the most represented across government functions 74

Figure 3.1. Most European Union (EU) AI use cases are in pilot or development phases 82

Figure 4.1. OECD Open, Useful and Re-usable data (OURdata) Index, 2023 115

Figure 4.2. Data governance in the public sector 116

Figure 4.3. Considering the level of AI literacy needed for different user groups in the workforce 122

Figure 4.4. OECD GovTech Policy Framework 131

Figure 4.5. Key steps to understand the users and their needs in government AI developments 144

Figure 4.6. OECD Framework for Trustworthy Artificial Intelligence in Government 145

Figure 5.1. AI deployments across OECD Members who use AI in tax administration 166

Figure 5.2. Objectives for FMIS upgrades in OECD countries, 2022 179

Figure 5.3. Potential use of AI and data analytics throughout the public procurement cycle 201

Boxes 19

Box 1.1. The evolution of AI 19

Box 1.2. The EU AI Act's risk levels as related to AI in government 33

Box 1.3. The concept of "high-impact" AI in United States (US) policy 34

Box 3.1. Government AI lab's experience using proprietary AI 92

Box 4.1. Accelerating federal government use of AI in the United States 107

Box 4.2. Artificial Intelligence Playbook for the UK Government 109

Box 4.3. France's personal data sandbox for AI in public services 113

Box 4.4. Sweden's collaborative data gathering for Svea 114

Box 4.5. Korea's shared data centres and government cloud 119

Box 4.6. European Union competency framework for AI in government 123

Box 4.7. Driving efficient AI acquisitions in the United States 129

Box 4.8. Public sector algorithmic auditing approaches and tools 139

Box 5.1. Detecting undeclared property development in France 167

Box 5.2. Improving decision-making in Brazil 167

Box 5.3. Improving compliance checks with predictive analytics in Austria 168

Box 5.4. Enhanced taxpayer services in Singapore 169

Box 5.5. Preventing errors in Australian tax returns 169

Box 5.6. Challenges with algorithmic errors in tax administration 171

Box 5.7. Forecasting GDP with explainable AI in Sweden 173

Box 5.8. AI for public financial management in Korea 174

Box 5.9. France uses AI in budget monitoring 175

Box 5.10. AI-driven fiscal transparency in Brazil 176

Box 5.11. The Robodebt scheme: Challenges with collecting improper payments 177

Box 5.12. Using AI for legislative drafting and responses in the United Kingdom 183

Box 5.13. Developing AI for measuring regulatory compliance costs in Germany 184

Box 5.14. Denmark's SAFE AI tool 186

Box 5.15. Risk criteria for agricultural contributions in the Autonomous Province of Trento 186

Box 5.16. Tuscany and the OECD develop risk criteria for grant applications 187

Box 5.17. Enhancing risk criteria through customer complaints in the Lazio region 188

Box 5.18. Good standing approvals and compliance in Israel 188

Box 5.19. France's strategy for using AI in government human resource management 193

Box 5.20. AI-enabled recruitment automation in the United Kingdom 194

Box 5.21. Increasing representation of visible monitories in Canadian defence leadership 195

Box 5.22. Generating learning content in Australia 196

Box 5.23. Ukraine's ProZorro e-Procurement System 202

Box 5.24. Chile advances public procurement with ChileCompra 203

Box 5.25. Brazil's AI-powered procurement oversight with Alice 204

Box 5.26. GovAI Coalition for responsible AI procurement and deployment in the United States 206

Box 5.27. Europe's DATACROS tool for detecting corporate ownership anomalies 211

Box 5.28. Using AI to tackle benefit fraud in the United Kingdom 211

Box 5.29. Brazil enhances audit processes through generative AI with ChatTCU 213

Box 5.30. Corruption prevention through the analysis of legal drafts in Lithuania 214

Box 5.31. Norway uses text mining and ML in police audits 219

Box 5.32. Evaluating the effectiveness of environmental policies with AI 220

Box 5.33. European Commission's eSummary tools to support public administrations 221

Box 5.34. Retrieval-Augmented Generation for LLMs 221

Box 5.35. Democratising access to the European Parliament's archives 226

Box 5.36. AI as facilitator of deliberative processes 227

Box 5.37. Improving participation in France's Grand Débat National 228

Box 5.38. Finland uses generative AI for participatory urban planning 229

Box 5.39. Innovating public communications with AI in the United Kingdom 229

Box 5.40. Practices of AI for information curation in citizen participation 230

Box 5.41. Chatbots supporting participation in Spain and Colombia 230

Box 5.42. AI solutions in planning processes 236

Box 5.43. Automation for smoother delivery of services in Sweden 237

Box 5.44. AI-powered visitor flow forecasting with Tampere Pulse in Finland 237

Box 5.45. Portugal's Land Cover Monitoring System (SMOS) 238

Box 5.46. Chatbots that streamline access to information 239

Box 5.47. United Arab Emirates Smart Mission for consular services 240

Box 5.48. France's AI-assistance for personalised public service delivery 242

Box 5.49. How social protection systems are embracing AI 242

Box 5.50. PES in OECD countries are increasingly turning to AI for service provision 243

Box 5.51. Korea's AI-powered multilingual healthcare communication 244

Box 5.52. How cities are using AI tools to provide services 245

Box 5.53. Germany's risk assessment tool for Islamist terrorism 252

Box 5.54. Anticipating wildfire risks in Canada 253

Box 5.55. AI and satellite images processing to protect against wildfires in Brazil 254

Box 5.56. Korea's AI-powered CCTV to protect against crime and emergencies 255

Box 5.57. Singapore's data-driven virtual twin 255

Box 5.58. The United States' hazard mitigation planning assistant 256

Box 5.59. Accelerating post-hurricane recovery with AI in the United States 257

Box 5.60. Facial recognition privacy and bias concerns 259

Box 5.61. Enhancing document management in Spain 263

Box 5.62. Prometea and ChatGPT to make Argentina's judicial sector more efficient 264

Box 5.63. Redesigning justice services with AI for effective protection measures in Peru 265

Box 5.64. AI-powered anonymisation of court decisions 266

Box 5.65. Portugal's AI-powered chatbot to provide practical information 267

Box 5.66. Chatbot supports child victims with tailored legal information 267

Box 5.67. Risk assessment system for domestic violence in Spain 268

Box 5.68. Harnessing AI oversight in Brazil's Supreme Court 269

Box 5.69. Spain: National Policy on the Use of AI in the Administration of Justice 271

Box 5.70. Judicial guidance on AI in the United Kingdom 273

Box 5.71. Colombia's directive on AI rules and guidelines for the judiciary 273