While key enablers for AI adoption in government are reaching maturity, delivery capabilities remain uneven. Governments are taking important steps to upskill public servants, but there is room for growth in using AI for specific purposes. And while most countries fund AI initiatives, support for procurement lags. Cloud computing capacities for AI are also solidifying but other forms of digital infrastructure are less developed. Strong governance is the cornerstone of successful AI adoption in government, and most countries have oversight or advisory bodies, but activity focuses on guidance rather than enforcement. Similarly, most countries commit to algorithmic transparency but few have formal standards or open algorithm registers. In addition, limited internal repositories of AI use cases constrain transparency and governance. Challenges to measuring the impact of AI limit decision-making and contribute to a proliferation of pilots with little potential to scale. While stakeholder engagement around strategies is strong overall, sustained, user and cross-border involvement remain limited.
Digital Government Outlook 2026
From Foundations to Transformational Impact
Abstract
Artificial intelligence (AI) use in government is spreading, but not evenly. AI is now used in at least one area of government in 35 of 36 (97%) of OECD countries, with strongest uptake in internal processes and public services. Use in policymaking and oversight remains more limited, reflecting higher-stakes applications and stronger requirements for data quality, transparency and assurance.
Governance frameworks for AI in government are becoming more widespread, but enabling conditions remain uneven. Nearly all OECD countries have an AI-in-government strategy and 30 of 36 countries (83%) have at least one institution responsible for governing AI in the public sector. Yet the practical conditions for scaling AI – including data governance, infrastructure, skills and organisational capacity – remain uneven across countries.
AI skills efforts are expanding, but activity-specific training lags. Across the OECD, 32 of 36 countries (89%) report training programmes to support AI in government. Yet fewer offer training on using AI in public services or policymaking (13 of 36 countries, or 36%, each), suggesting that broad capability is improving faster than role-specific readiness.
Funding outpaces procurement readiness. Most OECD countries fund AI initiatives, but only 21 of 36 countries (58%) provide central support for procuring AI goods and services. Governments therefore need stronger capabilities to manage vendor lock-in, accountability, transparency, data rights and lifecycle risks.
Guardrails are expanding, but operational controls lag. All OECD countries have at least one form of guardrail, yet only 14 of 36 countries (39%) require pre-deployment risk assessments, 12 of 36 (33%) have internal review committees and 11 of 36 (31%) conduct post-deployment audits. Transparency is also weak: only 11 of 36 countries (31%) have a formal standard and 6 of 36 (17%) have an open algorithm register.
Evidence and user feedback remain limited. Only 10 of 36 OECD countries (28%) report any financial or non-financial impact measurement of AI use cases in government, even though half say adoption decisions draw on evidence of potential efficiency or cost savings. Engagement is strong in strategy development, but much weaker in implementation: only 15 of 36 countries (42%) engage service users and 8 of 36 (22%) have citizen feedback or complaint mechanisms.
Artificial intelligence (AI) is among the most transformative forces of the 21st century, becoming a critical component of the digital government landscape. Across OECD countries, use of AI in government is shown to improve government productivity (efficiency and effectiveness), support more proactive and human-centred services, and strengthen responsiveness and accountability.1 As governments confront increasingly complex policy challenges, AI can play a central role in improving decision making, automating routine processes and scaling integrated services.
Unlocking AI’s benefits depends on the strength and resilience of underlying digital systems and surrounding ecosystems. As highlighted in previous chapters, AI adoption is only as strong as the foundations on which it rests, including high-quality and interoperable data, coherent digital public infrastructure (DPI), adaptive investment and procurement models, and a public sector workforce with the skills to develop, govern and oversee AI. Where these foundations are weak or fragmented, AI cannot scale effectively and might amplify rather than mitigate risks.
Managing AI’s risks is therefore integral to advancing trustworthy and resilient digital transformation. Governments must mitigate ethical risks that can create adverse outcomes and rights infringements, operational risks that erode trust, exclusion risks that widen digital divides, and public resistance to the use of AI by governments. Not adopting AI also presents a risk: it can lead to missed opportunities to enhance services, improve efficiency and strengthen evidence-informed policymaking.
However, governments face implementation challenges, including skills shortages, outdated legacy systems, inadequate data governance and fragmented investment frameworks. Public administrations also need to pursue AI in ways that reinforce the public interest, uphold rights and ensure societal benefits to distinguish government use of AI from private sector incentives.
The OECD flagship report Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions highlights that advancing AI in government requires a structured and holistic approach. The OECD Framework for Trustworthy AI in Government (Figure 4.1), provides a roadmap to help governments align their initiatives with the OECD AI Principles (2024[1]). The Framework organises specific activities around three pillars: (1) enablers to facilitate adoption; (2) guardrails to guide trustworthy use; and (3) engagement approaches to shape user-centred and responsive adoption.
Source: (OECD, 2025[2]).
Across the OECD, government AI maturity continues to improve, reflecting efforts to build the enablers, guardrails and transparency mechanisms needed for responsible AI adoption. The OECD Digital Government Index (DGI) AI maturity component assesses how well central governments are prepared to use AI strategically and responsibly. This component provides scores to questions in the OECD Survey on Digital Government 3.0, covering enablers such as national AI-in-government strategies and the openness of engagement of their design, guardrails such as oversight and ethical advisory bodies, transparency in the use of algorithms, and the extent to which AI is used in government operations, policymaking and public services. To complement these indicators, the Survey includes a separate AI Annex with qualitative questions to inform the narrative analysis in this chapter. Responses to the Annex were not scored for the 2025 DGI but provide contextual evidence and illustrative examples (see the Methodology annex). For the 2025 DGI, no data are available for Germany and the United States because they did not participate in the Survey.
Between 2023 and 2025, most OECD countries either improved or maintained their government AI maturity. Estonia, France, Korea and the United Kingdom remain consistently high performers, driven by the consolidation of ethical or regulatory oversight bodies and the implementation or strengthening of transparency mechanisms, particularly algorithmic transparency instruments. France is notable for having in place all items measured in the AI maturity component. Belgium, Israel, Latvia, Norway, and Portugal made significant progress compared to 2023 baselines, largely in the adoption of new strategies, the expansion of AI use and more open engagement approaches. Many other countries recorded smaller improvements, and a few saw declines indicating that maintaining momentum in implementation is as important as introducing new instruments.
Looking ahead, European Union (EU) Members might be poised to advance further as the requirements of the EU AI Act (2024[3]) take effect. Although the Act was adopted in 2024, many provisions relevant to the DGI, such as for algorithmic transparency mechanisms, were not yet applicable during the 2025 DGI analysis window. While a small number of EU Members took proactive steps in advance toward compliance, most had not yet implemented the Act’s obligations at the time of measurement. This suggests a likely future increase in maturity scores as countries align their practices with forthcoming requirements.
Taken together, these developments highlight that OECD countries are making steady progress in building the foundations for trustworthy and strategic AI use. However, they also reinforce a recurring theme of the report: governance instruments must be actively implemented, continuously maintained, and integrated into day-to-day decision making to deliver sustained improvements in AI maturity and digital resilience.
AI adoption in government has expanded significantly, particularly for internal processes, but remains uneven across government activities. The 2025 DGI analysis window coincided with a surge of interest in AI, fuelled by rapid advances in generative AI (GenAI). During this period, the share of OECD countries using AI for internal processes rose from 23 of 33 measured countries (70%) in 2023 to 31 of 36 (86%) in 2025, and adoption in public services increased from 22 of 33 countries (67%) to 27 of 36 (75%) (Figure 4.2). However, adoption remains more limited in policymaking and accountability activities. Only 13 of 36 OECD countries (36%) report using AI to support policymaking (up from 11 of 33 countries, or 33%, in 2023), and 12 of 36 countries (33%) have used AI to strengthen oversight and accountability.
To some extent, the gains reflected in internal processes and public services reflect the relative ease of applying AI to structured administrative tasks – such as document classification and workflow optimisation – compared to policymaking and accountability activities, which can be higher-stakes, involve more contestable judgements, and often have more complex governance and data requirements. More broadly, uneven uptake across government activities reflects varying constraints depending on the type of activity – such as skills shortages, legacy IT, and difficulties accessing and sharing high-quality data – alongside greater requirements for privacy, transparency and representation that can be more demanding in policy and oversight settings.
As a result, AI use in policymaking and accountability activities is often narrower and more cautious, and depends on more comprehensive, integrated data and stronger assurance and oversight arrangements than many administrative applications. Furthermore, these conditions remain uneven across countries (Chapter 2).
Overall, this report shows that AI use expands most rapidly where foundations are strong, and more slowly where risks, data gaps or governance constraints are greatest. This mirrors other OECD (2025[2]) work specific to AI in government, which found that uptake tends to be faster where data are available and processes are more standardised, and slower where legacy systems, skills gaps, and higher requirements for privacy, transparency and representation raise the bar for deployment.
Still, AI use is nearly universal across the OECD area, with 35 of 36 OECD countries (97%) and three of six accession candidate countries (50%) using AI in at least one of these activities. Seven OECD countries (Chile, Estonia, France, Korea, Latvia, Luxembourg and Norway) use AI in all four areas, while no accession candidate country has reached this breadth.
Box 4.1 offers notable examples of AI applied across OECD countries, including cases that illustrate how countries move from experimentation toward operational deployment. Additional examples from beyond the 2023/24 analysis window are documented in the OECD (2025[2]) report Governing with AI and listed on the OECD.AI Policy Navigator,2 highlighting the rapid pace of change in this field.
Percentage of OECD countries using AI in the public sector by policy area, 2023 and 2025
Note: Oversight and accountability in the public sector was not measured in the 2023 survey, therefore no comparison is available. Accession candidate countries report the following use rates: Internal processes (25%); Public services (38%); Policymaking (0%); Oversight and accountability (17%). 2025 data not available for Germany and the United States. 2023 data not available for Bulgaria, Germany, Greece, Indonesia, Slovak Republic, Switzerland, Thailand and the United States. Data for Indonesia and Thailand cover 1 January 2022 to 31 December 2023, while Oversight and accountability does not include data for these countries. Refer to Annex Table 4.A.1 for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
AI increasingly demonstrates relevance for a range of government activities, including:
Internal processes, such as Korea’s e-RFP Assistance System, which brings together several types of AI, including GenAI, to draft requests for proposals and ensure compliance with regulations. The system provides a repository of past government procurement data to inform government officials. Its adoption has resulted in a 70% reduction in document preparation time compared to traditional approaches, and 99.9% compliance with procurement regulations.
Public services, such as Iceland’s Askur AI chatbot, which supports citizens in accessing information and conducting services on the government’s central portal. Askur addresses 90% of citizen correspondence, significantly reducing phone calls and emails to service centres.
Policymaking, such as the United Kingdom’s Parlex is a research assistant for parliamentary information. It can quickly find and analyse key contributions from members of parliament, parliamentary questions, and committee hearings. Parlex helps bring the views of parliamentarians directly into the policymaking process, prepare ministers more effectively, and refine policy before it is introduced.
Oversight and accountability, such as Brazil’s Alice system, which continuously analyses public procurement activities for potential risks of fraud, errors and inefficiencies, and facilitates interventions when they are detected. In 2023, it analysed nearly 191 000 acquisitions and triggered 203 audits involving contracts worth EUR 4.15 billion (equivalent). The system also enabled a reduction in audit processing from 400 days to 8 days.
While use of GenAI in government is expanding, it remains less prevalent than the use of broader AI and machine learning (ML) systems. This is consistent with findings from the OECD (2025[2]) and the European Commission (EC) (2025[9]) that show uneven levels of preparedness for, and use of GenAI systems in government across countries. Despite these challenges, 64% of OECD countries and 50% of accession candidate countries use GenAI systems for at least one purpose.
Among OECD countries, GenAI use tends to concentrate in lower-risk, productivity-enhancing applications rather than in high-stakes operational contexts. The most common uses include: supporting public servants in their functions (20 of 36 countries, or 56%), such as through drafting assistance, search and knowledge retrieval, or conversational support; generating automated reports or summaries to inform decision-making (15 of 36 countries, or 42%); creating citizen-engagement content (15 of 36, or 42%), such as consultation replies and newsletter texts; drafting or assisting in the creation of policy documents (13 of 36, or 36%); and supporting the design and delivery of public services (13 of 36, or 36%), most often during early ideation, user research or prototype-development phases.
These uses reflect a cautious approach that prioritises internal productivity and augmenting human work over fully automated processes or decision making. They also illustrate reliance on commercial GenAI products such as Microsoft Copilot, OpenAI’s ChatGPT or Mistral’s Le Chat, and on self-hosted pre-trained open-source or open-weight systems such as Meta’s Llama models. Open-weight systems make a model’s trained parameters (“weights”) available for others to run and fine-tune even if the training data and full development process are not openly shared (OECD, 2025[10]). Governments may favour open-weight systems because they can be deployed in contained environments, offering greater control over data handling and security, more scope for customisation to local languages and procedures, and reduced dependence on a single vendor.
Notably, no OECD Member or accession candidate country imposes a blanket ban on the use of GenAI systems in government, although some restrict the use of specific systems such as DeepSeek and other high-risk or unverified tools. This reinforces a broader trend observed throughout this chapter: governments are generally open to exploring GenAI, but adoption remains measured and risk sensitive, shaped by concerns around security, data protection, quality assurance and public trust.
Enablers are the foundational elements necessary for trustworthy and scalable AI implementation in government. The OECD identifies seven enablers: governance; data (including open government data); digital infrastructure; skills and talent; AI investment; public procurement; and partnering with non-governmental actors (2025[2]). Several of these – notably data, infrastructure, investment and partnering – are not AI-specific but structural conditions that support all digital transformation. As highlighted in the preceding chapters, these foundations are prerequisites for digital resilience, and their uneven maturity shapes governments’ ability to adopt AI safely, strategically and at scale.
4.4.1. Strong governance is the cornerstone of successful AI adoption in government
Robust governance is the primary enabler for trustworthy and effective AI use in government. Two components are critical: (1) a high-level national strategy that articulates objectives for adopting AI in government; and (2) a whole-of-government approach to co-ordinating its implementation, ensuring coherence across ministries and establishing clear accountability for results.
Nearly all OECD countries have such a strategy. During the 2025 DGI analysis window, all but three OECD Members (Canada, Mexico and Switzerland) had an AI-in-government strategy. By 2025, Canada and Switzerland published strategies, marking a small improvement since the 2023 DGI. Among accession candidate countries, Argentina and Brazil also have national strategies.
These strategies commonly emphasise trustworthy, human-centric AI, highlighting principles such as fairness, accountability, robustness and security, in line with the OECD AI Principles. They often position government as a testbed for innovation while setting priorities that mirror the broader digital-government agenda. Common priorities include building data and digital infrastructure, strengthening public sector skills and organisational capabilities, creating dedicated co-ordination and governance mechanisms, and promoting experimentation through pilots, sandboxes and cross-sector collaboration.
Importantly, most countries designate institutions to implement these strategies. According to the DGI, 30 of 36 OECD countries (83%), and four of six accession candidate countries (67%) have at least one public institution responsible for governing the use of AI in the public sector (Figure 4.3). These bodies play central roles in shaping standards, ensuring ethical oversight, guiding procurement and risk management practices, and helping translate high‑level intentions into operational decisions (Box 4.2).
Type of institution(s) responsible for governing AI in the public sector, by country, 2025
Note: Data not available for Germany, the United States, Indonesia or Thailand. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
Spain
The Spanish Agency for the Supervision of AI (AESIA) is the public body responsible for ensuring the ethical and safe use of AI by public and private entities in Spain, as mandated by the EU AI Act’s requirement for national competent authorities. This includes drafting guidelines and legislation for the use of AI in public services, which ensure that: civil servants are sufficiently trained; fundamental rights of citizens are sufficiently protected; and projects comply with national and EU legislation around data governance, transparency and robustness for AI systems.
Czechia
In Czechia, the Committee for AI serves as a permanent advisory and co-ordinating body. Its mandate is to support the implementation of the National Artificial Intelligence Strategy 2030, which outlines “AI in public administration and public services” as one of seven key areas. The committee fulfils this function by publishing regularly updated Action Plans that put forward projects and establish key performance indicators for various goals. These goals include expanding the use of AI in the public sector and ensuring that public-sector employees are aware of the possibilities and limitations of AI.
Source: (Agencia Española de Supervisión de la Inteligencia Artificial (AESIA), n.d.[11]; Agencia Española de Supervisión de la Inteligencia Artificial (AESIA), n.d.[12]; Agencia Española de Supervisión de la Inteligencia Artificial (AESIA), n.d.[13]; Ministry of Industry and Trade, Czech Republic, 2024[14])
4.4.2. Governments take important steps to upskill public servants, with room for growth in using AI for specific purposes
Strong governance frameworks for AI must be matched by the skills and confidence of public servants who use AI tools in their daily work. While strategies and central co-ordination provide essential direction, it is frontline public servants who ultimately make decisions about how AI is applied, interpreted and overseen. Consistent with earlier chapters, the OECD finds that the skills gaps remain the most common challenge hindering successful AI adoption in government (2025[2]).
To close these gaps, 32 of 36 OECD countries (89%) and three of six accession candidate countries (50%) have training programmes for various skills required to support the development and use of AI in government. These focus most frequently on the practical and ethical use of AI tools (28 of 36, or 78%, and 22 of 36, or 61%, of OECD countries, respectively), and data privacy and security (20 of 36 countries, or 56%), with fewer offerings for using AI in public services or policymaking (each 13 of 36 countries, or 36%) (Figure 4.4 and Box 4.3).
Training initiatives typically fall into a few recurring patterns, often combining broad baseline literacy with role-specific depth:
Broad foundational courses. Many countries make widely available online courses, such as the globally accessible “Elements of AI” course, accessible through civil-service learning portals (e.g. Czechia, Luxembourg and Norway), or generalist training for the French public service designed by the Digital Campus and accessible through the MENTOR platform, to build baseline awareness of opportunities and risks.
Government-tailored fundamentals. Some administrations offer introductory training aligned with domestic frameworks and guidance, such as Australia’s “AI in Government Fundamentals” course for the Australian Public Service (APS), which emphasises safe and responsible use consistent with Australia’s AI Ethics Principles.
Specialised technical training. Some programmes target specific capabilities and functions and may require longer participation or in-person engagement, such as the Korean Internet and Security Agency’s (KISA) “AI Security Control” course on security monitoring and response to AI-enabled attacks (with both extended in-person and shorter online options).
Many initiatives were launched after 2024, helping close gaps in several countries while bolstering efforts in those with existing offerings:
Australia’s APS AI Plan, which sets expectations for mandatory capability development and training support for safe and responsible use of GenAI tools;
Colombia’s 80-hour, fully virtual and free AI diploma for public servants, delivered with the University of Cartagena and designed to reach thousands of officials;
Costa Rica’s national partnership with UNESCO to equip public servants with AI and digital transformation skills, including an online course on AI and digital transformation in government;
Finland’s new state-administration learning packages that include an “AI academy” video curriculum within broader digital skills provision;
Lithuania’s AI training platform initiative for public sector employees, with staged training and materials covering practical use, privacy, ethics and EU AI Act-related topics;
the United Kingdom’s One Big Thing “AI for All” initiative, intended to build confidence and responsible use of AI across the civil service.
Brazil's AI training portal for public servants offers structured learning pathways tailored to five professional profiles (from frontline staff to senior leaders) covering applied AI, data governance, ethics and strategic decision-making.
Source: (University of Helsinki and MinnaLearn, n.d.[15]; Australian Public Service Academy, n.d.[16]; Korea Internet & Security Agency (KISA), n.d.[17]; UK Cabinet Office, 2025[18]; Australian Department of Finance, 2025[19]; Ministry of Information Technologies and Communications (MinTIC), 2025[20]; UNESCO, 2025[21]; Ministry of the Economy and Innovation, 2025[22])
Percentage of OECD countries reporting training programmes to support AI skills, by topic, 2025
Note: Data not available for Germany, the United States. Data cover 1 January 2023 to 31 December 2024. Refer to Annex Table 4.A.2 for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
4.4.3. Most countries fund AI initiatives, but support for procurement lags
Financial resources are as essential as human capabilities for enabling AI adoption, but funding and financing mechanisms often receive limited attention in national strategies for AI in government. Despite this, most OECD countries take steps to resource government AI initiatives. According to the 2025 DGI, 32 of 36 OECD countries (89%) (all except Costa Rica, Hungary, Mexico, and Switzerland) and four of six accession candidate countries (67%) (all except Croatia and Romania) have some form of funding for the development or use of AI systems in government. Among OECD countries that report having funding mechanisms for AI, 15 of 32 countries (47%) have dedicated funding, while 17 of 32 (53%) rely on broader digital government funding streams.
Investments take various forms, including public-sector incubators, dedicated grants, and targeted support programmes to develop or test AI systems (Box 4.4). Despite different institutional designs, all three of these mechanisms channel funding toward practical AI experimentation and delivery, not just strategy or research. They are designed to reduce barriers to early adoption, lower the risks of experimentation for ministries and agencies, accelerate real‑world testing, and provide structured support for scaling. This signals a broader trend: governments see AI funding as a way to convert institutional interest into tangible implementation, especially given the constraints of traditional budgeting cycles.
However, funding alone is insufficient if governments lack support for procuring AI technologies. Only 21 of 36 OECD countries (58%) and no accession candidate country provide central government support for procuring AI goods and services. Given the complexities of AI systems and the inherent challenges within public procurement processes, these modest levels of support warrant further action. Without strong procurement guidance, innovation procurement procedures and shared mechanisms such as standard criteria, frameworks and model contract clauses, governments risk vendor lock-in, unclear accountability, and limited transparency into system behaviour and performance. Contracts may also fail to adequately address risk, licensing, data and intellectual property rights, auditability, monitoring, or model lifecycle management and exit arrangements.
Governments therefore need strong procurement capabilities if they want to use AI well. Buying AI systems is not just about choosing a vendor. It requires enough technical understanding to define what the system should do, set clear safeguards, and check whether suppliers are handling data responsibly, building reliable models, keeping systems secure, and delivering results over time. These capabilities are essential if governments want to move beyond small pilots and use AI in ways that are safe, efficient, and responsive to citizen needs.
As explored in Chapter 3, governments’ ability to engage with non-governmental actors through public procurement, public-private partnerships or collaborative innovation models is an important enabler of broader digital transformation efforts. AI further magnifies this need: governments must be able to partner strategically, outsource selectively, and retain sufficient internal capability to govern technology responsibly.
France
France’s ALLianNCE incubator supports government actors in AI adoption with a community of learning, guidance, and funding for talent acquisition averaging EUR 100 000 per project. This approach enables agencies to bring in specialised expertise to design and deploy AI systems. The impact of this talent‑oriented funding model is reflected in the eight AI products incubated in 2024. Co-financed by the initiatives, the resulting projects addressed opportunities such as automated transcription and French-language LLM calibration, and supported several agencies in the central government.
Denmark
Denmark created two technology investment funds between 2020 and 2024 to encourage integration of AI systems into the public sector. From municipal to central authorities, multiple levels of government were eligible to receive financing. The Signaturprojekterne fund encouraged development of AI focused on building concrete public sector experience with the new technology. The Tilskudspulje for nye teknologier fund supported new technologies that might address societal challenges. The former supported development of 40 AI projects with approximately EUR 26 million in total funding, and the latter has seen several AI projects awarded.
Türkiye
Türkiye’s Scientific and Technological Research Council supports the procurement and co-development of AI through funding consortia. Its efforts bring together public institutions as end-users, and tech firms or research organisations as developers, enabling AI systems to be tailored through structured R&D collaborations. The programme prioritises five domains: (1) smart manufacturing systems; (2) smart agriculture and food; (3) financial technologies; (4) climate change and sustainability; and (5) smart education technologies. The programme is a central mechanism for strengthening Türkiye’s AI ecosystem with 41 projects with approximately EUR 4 million over three years.
4.4.4. Cloud computing capacities for AI are solidifying but other forms of digital infrastructure are less developed
Digital infrastructure is critical for AI development in government, acting as the connective tissue that enables systems to scale, interact and operate reliably (Chapter 2). Figure 4.5 and Annex Table 4.A.3 present AI-relevant infrastructure in OECD Members and accession candidate countries. The 2025 DGI findings show OECD countries making progress, particularly in cloud capabilities, but other components remain less mature.
Most OECD countries (26 of 36 countries, or 72%) have cloud computing capacity to support AI. This is an encouraging sign of progress toward scalable, flexible and resilient infrastructure, especially given the compute intensity of many modern AI systems. Depending on the arrangement, cloud environments allow governments to access elastic capacity, advanced tools, and managed services without the long lead times associated with on‑premise infrastructure.
However, only 13 of 36 OECD national governments (36%) report using hardware accelerators such as Graphics Processing Units (GPUs). This likely reflects a mix of factors: greater reliance on cloud or managed AI services, where accelerator capacity is abstracted away from public organisations; the fact that many AI use cases do not require dedicated accelerators; and barriers to acquiring and operating specialised hardware, such as long procurement lead times and supply-chain bottlenecks for advanced accelerators, skills gaps and the operational requirements of hosting them securely.
More concerning is the low maturity of data governance frameworks and the unavailability of high-quality data for AI training. The gaps echo findings from Chapter 2: many governments continue to face fragmented data ecosystems, weak quality management, and limited reuse of authoritative datasets. Without strong data foundations, AI models cannot perform reliably and increase risks related to skewed data, poor accuracy, and unreliable outputs.
Adoption of privacy-enhancing technologies (PETs) is also low. While this is unsurprising given their relative novelty and specialised implementation demands, it represents an area for future investment. PETs can help operationalise privacy-by-design principles, support data minimisation, enable responsible data sharing, and facilitate cross-border collaboration for training and evaluation while safeguarding privacy, confidentiality and intellectual property rights (OECD, 2025[2]; 2024[28]; 2023[29]; OECD, 2025[30]).3 As AI systems handle more sensitive information, and as interoperability expands across organisations and borders, PETs will become increasingly important for maintaining public trust and ensuring compliance with legal and policy requirements.
Box 4.5 presents some promising practices identified during the 2025 DGI analysis. They show that AI‑ready governments do not rely on a single infrastructure solution, but build layered, interoperable, and resilient approaches that combine cloud capacity, compute, governance, data accessibility, security and shared AI services. These examples indicate a clear shift from siloed infrastructure to shared, government‑wide platforms, and a combination of commercial and sovereign approaches recognising that no single infrastructure model meets all needs.
Percentage of OECD countries having selected digital infrastructures and components to support AI integration, 2025
Note: Data not available for Germany and the United States. Data cover 1 January 2023 to 31 December 2024. Refer to Annex Table 4.A.3 for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
Governments around the world undertook a range of investments to strengthen their AI-enabling digital infrastructure. For example:
Cloud computing platforms. Belgium’s G-Cloud is a hybrid cloud, using services both offered by private companies in public cloud environments and those hosted in government data centres. The government manages the G-Cloud while the private sector is responsible for its development and operations.
Data centres and storage systems. Austria’s Federal Computing Centre operates data centres and analytical infrastructure that process millions of records per year.
AI training resources. Portugal has over 2 000 standardised service factsheets available in open-dataset format, supporting consistent and accurate information, which have been used to train their citizen-facing ChatBot, a virtual assistant housed on the ePortugal Portal.
Hardware accelerators (e.g. GPUs). In 2024, the Export and Investment Fund of Denmark (EIFO) funded the Danish Centre for AI Innovation (DCAI), a new company that partially owns and will operate Gefion, Denmark’s first supercomputer. It is available to researchers from the public and private sectors.
Data governance frameworks. Brazil’s Data Maturity Model (MMD) is a tool for evaluating and improving data governance in public institutions. It is based on DAMA-DMBOK, a widely recognised international standard for data management, which adds credibility and structure.
Data integration and interoperability tools. Ena is Sweden’s national framework for digital infrastructure, led by the Agency for Digital Government (Digg). It supports data integration and interoperability by providing shared digital components, such as secure data exchange, standardised interfaces and common frameworks, that enable public-sector organisations to collaborate efficiently.
PETs. The United Kingdom Service Standard requires teams to follow secure-by-design approaches, including that they “collect, process and store data securely and in a way which respects users’ privacy”. The Office for National Statistics (ONS) publishes work on privacy-preserving synthetic data and discusses differential privacy to enable safer access to sensitive data. The government also explores privacy-preserving federated learning, where organisations train models locally and share model updates rather than raw data.
Self-deployed foundation models. ALIA is Spain’s publicly funded, open, public AI infrastructure, providing language models and related resources in Spanish and co-official languages (Catalan, Valencian, Basque and Galician), co-ordinated by the Barcelona Supercomputing Centre and intended for use by public administrations, researchers, universities and companies.
Countries such as Japan acted after 2024 to close gaps, while others took steps to bolster relatively strong positions. Several European Union Members enacted new legislation and policy in 2025 to comply with the EU Data Governance Act (DGA). While the DGA is not an infrastructure programme per se, it can strengthen foundations relevant to data infrastructure and interoperability by encouraging trusted data-sharing and reuse, including through secure processing arrangements. Use of PETs appears to be increasing slightly, with countries such as France and the United Kingdom conducting relevant efforts and Israel issuing guidance in 2025-2026.
Source: (Federal Public Service (FPS), Public Social Security Institutions (IPSS) and ICT organisations, n.d.[31]; Austrian Ministry of Finance (BMF), n.d.[32]; Republic Portuguesa, 2023[33]; Danish Centre for AI Innovation (DCAI), 2024[34]; Brazilian Secretariat of Digital Government, 2024[35]; Sweden Agency for Digital Government (DIGG), n.d.[36]; UK Government, n.d.[37]; Office for National Statistics (ONS) - Data Science Campus, 2023[38]; UK Department of Science, Innovation and Technology, 2023[39]; ALIA, n.d.[40]) (Israel Privacy Protection Authority, 2025[41]; LINC, 2025[42]; NHS Digital, n.d.[43]; UK Department for Science, Innovation and Technology, 2026[44])
Guardrails help ensure the trustworthy development, deployment and use of AI in government. They are essential for managing risks associated with AI and deploying AI according to legal boundaries and public values. Guardrails also support digital resilience by helping governments detect problems early, adjust implementation pathways and maintain public trust, all necessary for scaling AI safely.
However, they must be seen together with the enablers discussed earlier in the chapter. Strong guardrails without strong enablers can fuel risk-aversion and stall innovation. Likewise, strong enablers without adequate guardrails increase risk exposure. A balanced, proportionate approach, tailoring controls to the risk level of each use case, is critical to avoid both misuse and inaction (OECD, 2025[2]). Governments should determine which guardrails fit their operations and contexts, and apply them to AI uses in a manner commensurate and proportionate to their level of potential risk.
4.5.1. All OECD countries have high-level guardrails to ensure trustworthy AI
Across the OECD area, governments leverage a mix of formal requirements (e.g. binding regulations and mandatory standards) and soft policy levers (e.g. guidelines, standards, ethical principles) to ensure the trustworthy management and use of AI in government, in alignment with the OECD AI Principles. All OECD countries have at least one form of guardrail, 25 of 36 countries (69%) use formal requirements, 30 of 36 (83%) use soft approaches, and 19 of 36 (53%) use both. Among accession candidate countries, Peru uses both formal and soft approaches, while Argentina and Brazil rely primarily on soft mechanisms.
Within OECD countries, many guardrails come from broader digital and data-protection frameworks. For example, aspects of the EU General Data Protection Regulation (GDPR) operationalise transparency, fairness and accountability obligations relevant to AI.
The EU AI Act is also relevant. While most of its provisions were not yet applicable during the 2025 DGI analysis window, nor even as of early 2026, several EU countries took proactive steps to achieve compliance in advance. In addition to these regional obligations, countries have national principles and practices tailored to their contexts. For example, Greece released AI Working Guidelines as part of its national AI Strategy (Government of Greece, 2024[45]). Peru adopted dedicated AI legislation, demonstrating early regulatory action outside the EU. Countries outside the EU often use government-wide directives to adapt the OECD AI Principles to national priorities (Government of Peru, 2023[46]).
4.5.2. Implementation of guardrails remains limited, especially for enforceable controls
Despite continued progress at strategic and regulatory levels, implementation of guardrails remains uneven and often challenging. The 2025 DGI reveals that concrete, enforceable controls are far less widespread. Among OECD countries, 14 of 36 countries (39%) require ex-ante (pre-deployment) risk-assessments for AI systems, 12 of 36 (33%) have internal review committees overseeing AI use, and 11 of 36 (31%) conduct post-deployment audits (Figure 4.6). Box 4.6 provides examples of these initiatives.
The prevalence of ex-ante tools might reflect that they can be embedded in existing approval and procurement workflows, creating standard checkpoints, templates and sign-offs before deployment. However, ex-ante checks can be insufficient on their own. Guardrails also require clear, risk-based oversight mandates, continuous monitoring and well-designed audits that avoid creating false confidence or “audit washing” (OECD, 2025[2]). Review committees can add coherence and escalation routes across projects, but their impact depends on whether they can make or enforce decisions rather than operating as purely advisory bodies, and on having sufficient influence over government decision making. Ex-post auditing and ongoing monitoring are essential to detect drift (such as changes in data, behaviour or performance over time), and emerging issues and compliance gaps once systems are in use.
Many governments indicate that such measures are under development, including as part of their EU AI Act compliance efforts. Others have high-level principles and guidance in place, but lack concrete procedural requirements or operational processes to make these actionable. This gap reflects a broader pattern throughout this report: while governments make progress developing strong strategies and high-level approaches to digital transformation, translating them into routine practice through enforceable, risk‑proportionate mechanisms remains a significant challenge.
Without risk‑based assessments, audit structures, accountability frameworks and formal decision paths for high‑risk uses, guardrails might remain symbolic rather than operational. This weakens governments’ ability to detect emerging risks early, ensure adequate usage and accountability, govern vendor‑provided AI systems, uphold public trust and scale AI across institutions.
Many governments have concrete initiatives to facilitate the development and use of trustworthy AI systems in the public sector. Examples include:
Ex ante risk assessments. Canada’s Algorithmic Impact Assessment (AIA) is a mandatory risk-assessment being completed (with results openly published) by federal departments and agencies before deploying an automated decision-system. It verifies compliance with Canada’s Directive on Automated Decision Making and any other binding mandates. Another example, Australia’s AI Assurance Framework was piloted with 21 volunteer agencies from September to November 2024, testing a draft ex-ante AI impact-assessment tool to help teams evaluate AI use cases against Australia’s AI Ethics Principles, including benefits, reliability and risks.
Internal review committees. Luxembourg’s AI4Gov interministerial committee comprises representatives from the Ministry of Digitalisation, the Media and Communications Service (SMC), the Information and Press Service (SIP), and various technical experts. The committee aims to encourage government agencies to use AI and data science responsibly, to transform their actions and tasks, and to provide them with the necessary support.
Ex post auditing processes. As part of its National AI Strategy, Türkiye aims to establish an auditing mechanism to ensure the development of trustworthy and responsible AI. The country began a national risk-management certification programme.
Several governments took further action after 2024. Japan’s Council of Chief AI Officers was established in 2025 and facilitates risk-governance workflows, and the United Kingdom made updates to its Algorithmic Transparency Reporting Standard and refreshed its Data and AI Ethics Framework to include a self-assessment tool to identify impacts. Others sought to strengthen existing efforts. For instance, in late 2025, following its aforementioned pilot, Australia issued an updated policy for responsible AI in government, requiring agencies to complete the impact-assessment for certain AI use cases and establishing expectations for ongoing monitoring of AI use.
Source: (Government of Canada, 2026[47]; Türkiye Digital Transformation Office and Ministry of Industry and Technology, 2021[48]; Turkish Standards Institution (TSE), 2026[49]; Government of Luxembourg - Ministry of Digitalization, n.d.[50]; TSE Global, n.d.[51]; Digital Agency of Japan, 2025[52]; UK Government Digital Service, n.d.[53])
Percentage of OECD countries reporting controls to ensure trustworthy AI, by type, 2025
Note: Data not available for Germany and the United States. Data cover 1 January 2023 to 31 December 2024. Refer to Annex Table 4.A.4.for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
4.5.3. Most countries commit to algorithmic transparency but few have formal standards or open algorithm registers
Transparency around how government uses algorithmic AI systems and their outputs is important for building public trust (OECD, 2025[2]). Commitment to algorithmic transparency is growing across OECD countries, but mechanisms to operationalise it remain limited. In 2025, beyond their adherence to the OECD AI Principles, 21 of 36 OECD countries (58%) (up from 17 of 33 countries, or 52%, in 2023) and two of six accession candidate countries (33%) acknowledged the importance of algorithmic transparency as part of responsible AI use.
However, committing to transparency in principle is not the same as enabling it in practice. Only a minority of countries have instruments that make algorithmic transparency actionable. Two mechanisms matter most: (1) transparency standards and laws that require organisations to document how and why they use algorithmic tools; (2) open algorithm registers that publicly list algorithmic tools in use across governments. Across OECD countries, 11 of 36 countries (31%) have a law or other standard and 6 of 36 (17%) have an open register. Only six OECD countries have both (Canada, Estonia, France, Korea, the Netherlands, and the United Kingdom), while 25 of 36 countries (69%) have neither, meaning there is no formal structure to ensure transparency beyond high-level commitments. Box 4.7 presents examples of these mechanisms.
Among OECD accession candidate countries, only Peru has formal requirements in the form of a law to mandate the sharing of algorithm source code with political organisations, but only in the context of digital voting systems, making it narrow in scope (Congress of the Republic of Peru, 2025[54]).
Progress has been modest compared with 2023 DGI results, with the share of countries reporting either a transparency standard or algorithm register increasing from 7 of 33 countries (21%) to 11 of 36 (31%). Still, adoption remains low, suggesting that algorithmic transparency is a relatively weak area of AI governance across the OECD area. The gap between commitment and implementation again underscores the recurring theme of this report: governments recognise the importance of trustworthy AI but practical mechanisms for transparency, accountability and oversight remain underdeveloped. Strengthening algorithmic transparency will be essential to build public trust and support responsible AI scaling across the public sector.
Several governments take tangible steps to enhance transparency around the use of algorithmic systems in public services. Examples include:
Laws and guidelines. The European Union’s AI Act introduces extensive transparency obligations, with relevant requirements not yet applicable, and pending technical and governance requirements still not applicable during the 2025 DGI analysis window. Some EU Members took earlier action to mandate transparency, such as France’s Code des relations entre le public et l'administration, which requires the disclosure and explanation of algorithm-driven government decisions to impacted individuals. Outside the EU, Australia’s policy for the Responsible Use of AI in government requires transparency-driven actions in the adoption of AI from each agency, such as the publishing of a transparency statement explaining its overall use of AI and ongoing monitoring of systems for unintended impacts.
Algorithm registers. The government Algorithm Register in the Netherlands aggregates over 1 300 algorithms and provides details for each, such as its purpose and points of contact. The United Kingdom’s Algorithmic Transparency Recording Standard (ATRS) mandates public-sector organisations to disclose details about their use of algorithmic methods in decision-making, including who is responsible for the algorithm, its description and a breakdown of potential risks and mitigation activities. Currently, the ATRS lists 131 records.
4.5.4. Limited internal repositories of AI use cases constrain transparency and governance
An underlying challenge for trustworthy and scalable AI adoption is that most governments do not have a central repository of AI use cases. Repositories record critical information about AI systems, such as their objectives, implementation stages, outcomes, responsible institutions, timelines and underlying technologies. This information is important for enabling oversight, identifying duplication, monitoring risks, and building organisational learning across the administration. Figure 4.7 shows which countries have algorithmic transparency standards, central use case repositories, and open algorithm registers.
Only three OECD countries have mandatory AI use-case repositories (Australia, Canada and Estonia), while another 10 countries maintain repositories with optional contributions from agencies. No accession candidate country has such a repository. Estonia’s approach illustrates the value of a centralised repository. It documents almost 170 public-sector AI use cases across nearly 60 institutions (Government of Estonia, 2026[60]). Each record includes a description of the use case, the implementing institution, partners involved, the type of AI systems, area of impact and status details.
Several countries indicate either plans to create repositories or that they maintain partial lists tied to specific initiatives, but these do not provide a government-wide view. Such repositories can serve as a stepping stone towards an open register, and all countries with an open register have a central repository. The optional nature of most of these repositories raises questions about the completeness of the open registers.
Several countries acted after 2024 to close gaps. For instance, Colombia (Government of Colombia, 2025[61]; Government of Colombia, 2026[62]; Inspector General of Colombia and Ombudsman of Colombia, 2025[63]), Ireland (Government of Ireland, 2025[64]) and Sweden (IMY, 2025[65]) issued directives or guidelines to promote transparency for the public to understand how and why government uses AI. Japan put in place requirements for each ministry to compile an ongoing overview list of AI systems and report it regularly to the Digital Agency (Digital Agency of Japan, 2025[52]).
Countries reporting open AI registers, use case repositories and transparency standards, 2025
Note: Data not available for Germany, the United States, Indonesia or Thailand. All participating accession candidate countries responded “No” to all answer options. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
4.5.5. Most countries have oversight or advisory bodies, but activity focuses on guidance rather than enforcement
Oversight and advisory bodies are increasingly common in government AI governance ecosystems, but their functions remain oriented toward guidance and monitoring rather than hands‑on auditing or enforcement. In 2025, 30 of 36 OECD countries (83%) (and three of six accession candidate countries, or 50%) had either a regulatory oversight body or an ethical advisory body dedicated to AI. This represents an improvement since 2023, when 24 of 33 OECD countries (73%) had one of these bodies. Of OECD countries, 15 of 36 (42%) have both, as does Peru among accession candidate countries.
Among countries with such bodies, dominant activities include: developing procedural guidance such as guidelines or codes of conduct (27 of 30 countries, or 90%); producing technical or educational guidance (22 of 30, or 73%); and conducting ethical oversight and monitoring, such as through AI Councils or data-ethics bodies (20 of 30, or 67%).
More hands-on activities remain less common, including auditing, regulatory oversight and enforcement, or developing or executing reporting frameworks such as algorithmic impact-assessments. As one example of such of a regulatory oversight body, Portugal established a Specialised Monitoring Committee that serves as its national competent authority under the EU AI Act and ensures best practices in the development of Large Language Models (LLMs) and compliance with ethical rules.4 In contrast, an ethics advisory body makes non-binding recommendations, such as Japan’s Review Council for the Principles of Human-Centric AI Society, responsible for reviewing guidelines and codes of conduct in line with the ongoing societal impacts of AI (Cabinet Office of Japan, 2026[66]).
Several countries acted after 2024 to close gaps. For instance, Colombia (Consejo Nacional de Política Económica y Social, 2025[67]) and Japan (Digital Agency of Japan, 2025[52]; Digital Agency of Japan, 2025[68]) instituted advisory bodies comprised of internal and external experts, while Japan introduced an internal oversight body and oversight architecture.
Percentage of OECD countries with public bodies in charge of providing oversight or ethical advice for AI in the public sector, by type of oversight or advice provided, 2025
Note: 2025 data not available for Germany and the United States. Refer to Annex Table 4.A.5 for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
4.5.6. Guardrails for GenAI are less robust compared to other AI systems
Dedicated guardrails for generative AI (GenAI) remain comparatively underdeveloped across OECD countries, reflecting its more recent adoption in public administrations relative to other forms of AI. While governments have used AI for many years, GenAI has only been a focus for the last three to four years (Berryhill et al., 2019[69]; OECD, 2025[2]). Some research suggests that the pace of GenAI adoption outpaces the evolution of guardrails and training needed to use these tools responsibly (Giesecke, 2024[70]; Bright et al., 2024[71]).
GenAI in government differs from other types of AI and machine-learning (ML) systems. ML systems such as risk-scoring tools or resource-allocation models are typically bespoke, domain-specific and procured or developed through formal institutional processes, whereas GenAI is often built on general-purpose models accessible through chat interfaces (OECD, 2026[72]; 2025[2]). Public servants often have access to personal accounts for common GenAI systems and may leverage them in their functions, either with organisational approval or without as so-called “shadow AI”. Illustrating this pattern beyond government, while only 40% of companies report purchasing official LLM subscriptions, employees in 90% of surveyed firms report using personal AI tools for work (MIT NANDA, 2025[73]). In addition, GenAI’s ability to generate convincing but potentially inaccurate outputs creates distinctive integrity risks. Unlike many earlier AI systems that supported back-office decisions in the background, GenAI is often used to draft or edit citizen-facing content, policy documents and internal communications, which can be difficult to distinguish from human-authored material. This blurring of boundaries between human and machine-authored content raises questions about authenticity, accountability and trust, and underlines the need for appropriate, risk-based guardrails
In addition to (or as part of) the broader controls and transparency mechanisms discussed above, governments have guardrails specific to GenAI. To help ensure GenAI systems are used in a trustworthy manner, most OECD countries have ethical guidelines for the use of GenAI tools (27 of 36 countries, or 75%) and training programmes on the responsible use of such systems (21 of 36, or 58%). However, fewer have transparency measures in place, such as disclosure requirements for AI-generated content (13 of 36, or 36%); data standards or protocols to ensure security and privacy of data used or inserted into GenAI tools (13 of 36, or 36%); independent oversight by external bodies or experts (12 of 36, or 33%), or accountability frameworks to address potential misuse or errors (7 of 36, or 19%). Among accession candidate countries, two of six countries (33%) have training programmes and ethical guidelines, with countries lacking other guardrails (Box 4.8).
Examples of initiatives include:
Ethical guidelines. Chile’s 2021 National AI Policy was updated in 2023 following the accelerated advance of generative AI. Its reformulated third axis, Governance and Ethics, was published alongside relevant guidelines. Brazil’s Generative AI Primer in the Public Service guides public servants on the trustworthy use of GenAI tools and is updated as rules and technology evolve.
Training programmes. Australia's course on AI in Government Fundamentals includes a focus on when it is appropriate for public servants to use GenAI.
Transparency/disclosure requirements. The Netherlands’ Guide to the Responsible Use of Generative AI links use to compliance with domestic law, with requirements consolidated in the Algorithm Framework (including the Open Government Act). It also recommends maintaining documentation on when, why and by whom GenAI is used, publishing this via the Algorithm Register, supported by tools such as impact assessments and publication standards.
Data standards. New Zealand’s Responsible AI Guidance for the Public Service: GenAI clarifies data-handling and privacy controls for using GenAI, including assessing the sensitivity of prompts and any data shared with tools, and using Privacy Impact Assessments (PIAs) to check whether the proposed use aligns with privacy obligations and agency information-management requirements.
Accountability frameworks. Ireland’s Interim Guidelines for Use of AI in the Public Service place heavy emphasis on GenAI. They highlight the importance of human judgement, oversight and review.
Independent oversight. Under the EU AI Act, Members must designate a national competent authority to supervise and enforce the Act, including where public bodies use AI. In Spain, the Agency for the Supervision of AI (AESIA) supports this by publishing practical guidance to help organisations assess obligations under the AI Act, including for GenAI deployments and where systems might fall into higher-risk categories.
Source: (Chile's Ministry of Science, Technology, Knowledge and Innovation, 2024[74]; Chile's Ministry of Science, Technology, Knowledge and Innovation, 2023[75]; Secretariat of Digital Government of Brazil, 2025[76]; Government of the Netherlands, 2026[77]; New Zealand Digital Government, 2025[78]; New Zealand Government, n.d.[79]; Department of Public Expenditure NDP Delivery and Reform, 2024[80]; Spanish Agency for the Supervision of Artificial Intelligence (AESIA), n.d.[81]; Ministry of the Interior and Kingdom Relations, 2025[82])
4.5.7. Challenges in measuring the impact of AI limit decision-making and contribute to a proliferation of pilots with little potential to scale
A major barrier to strategic AI adoption in government is the widespread lack of processes to measure the financial and non-financial impact of government AI investments. The OECD (2025[2]) finds that inability to measure the value, outcomes or cost of AI systems is a core constraint on governments’ ability to make informed investment decisions. Without robust evidence on return-on-investment (ROI) or service impact, governments struggle to determine whether pilots are successful, justify scale-up or decide how to allocate limited resources.
This challenge is not unique to governments, as research on the private sector finds that “the main barriers to AI investment and adoption are a lack of understanding of AI benefits and the inability to measure them” (Ramos and Kandaswamy, 2023[83]). In contrast, companies with high levels of AI maturity succeed in delivering value because they design and implement structured metrics that quantify the benefits of their AI projects (Gartner, 2025[84]). Governments face similar difficulties, but with stronger accountability pressures and fewer organisational incentives to take risks.
Across OECD countries, impact measurement remains one of the least developed components of AI governance. Chapter 3 shows that almost half of OECD countries monitor broader digital investments, but only one in four evaluate their impact and realised benefits. These challenges also exist specific to AI investments: most governments do not have the processes for holistic measurement of potential or realised results of AI projects, such as efficiency of spend or quality of services (Figure 4.9). Only 10 of 36 OECD countries (28%) report conducting any financial or non-financial impact measurement studies of AI use cases in government, whether prospective or retrospective. Fewer still report measuring the impact of AI use across a government sector (4 of 36 countries, or 11%), or across government or how its use of AI affects society (6 of 36 each, or 17%).
Some governments initiated efforts to close these gaps. Australia (Australian Government, 2024[85]) and the United Kingdom (Government of the United Kingdom, 2025[86]) assess the impact of government-wide Microsoft Copilot trials, generating early insights into productivity and user experience. The United Kingdom also issued Guidance on the Impact Evaluation of AI Interventions in 2024 (updated July 2025), with tailored advice for applying the Treasury ministry’s Magenta Book to AI initiatives and helping teams understand whether, to what extent, how and why an AI intervention resulted in its intended impacts.
However, a mismatch persists between decision‑making and evidence. While only 28% of OECD countries report conducting an impact assessment on any AI use case, 50% report making decisions for adopting AI based on evidence of potential efficiency or cost savings. This discrepancy raises questions about exactly what evidence governments rely on, whether it is robust and comparable, and whether AI adoption is being guided more by expectations and pressure than by demonstrated value.
Recognising the need to address these challenges, the OECD is working on a practical guide to measuring AI impact in government to help policymakers and AI teams gauge impact at the project level and use the insights for decisions in organisations and across government portfolios (OECD, forthcoming[87]).
Percentage of OECD countries reporting measurement of the financial or non-financial impact of AI in government, by type, 2025
Note: Data not available for Germany, the United States, Indonesia or Thailand. Data cover 1 January 2023 to 31 December 2024. Refer to Annex Table 4.A.6 for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
Engaging stakeholders – including the public – is essential for building trust, legitimacy and accountability around how governments use AI. Public engagement helps ensure that AI systems reflect societal needs, reduce risks of exclusion, and support user‑centred design (OECD, 2024[88]). Rich engagement also lays the foundation for trustworthy and resilient AI governance, as communities understand how AI is used and have channels to shape its development.
4.6.1. Governments involve a range of stakeholders in their AI in government strategy development
One example of governments’ engagement around AI can be seen in the development of their national AI strategies. The 33 OECD countries with a strategy for AI in government were highly collaborative in developing these strategies. Among them, 30 of 33 countries (91%) engaged with academia, 29 of 33 (88%) with industry (large and/or established firms), 23 of 33 (70%) with civil society, and 22 of 33 (67%) with the GovTech community (start-ups and other SMEs) (Figure 4.10). The only potentially weak area was engagement targeting under-represented groups such as young people, women or indigenous communities (9 of 33, or 27%). In addition, 21 of 33 countries (64%) held open public consultations while developing their strategy, around the same as the 66% recorded in 2023. Australia, Chile, Estonia, Iceland, Ireland, New Zealand and Norway are noteworthy in that they engaged all five categories of stakeholders and held a public consultation. Among the two OECD accession candidate countries with strategies, Brazil held a public consultation and engaged with all categories except under-represented groups, while Argentina engaged with business, academia and the GovTech community.
Percentage of OECD countries indicating stakeholder involvement in the development of AI in government strategy, by stakeholder, 2025
Note: In addition to external engagement, all OECD Members except Türkiye, and all accession candidate countries report internal collaboration among public-sector organisations in developing the strategy. Data not available for Germany, the United States, Indonesia or Thailand. Refer to Annex Table 4.A.7. for comprehensive OECD and Accession country data.
Source: OECD (2025), Survey on Digital Government 3.0.
4.6.2. Governments engage beyond national strategies but ongoing user engagement and cross-border collaboration remain limited
Governments’ engagement around AI extends beyond strategy development, with many countries involving internal and external stakeholders in shaping public‑sector AI policies, use cases and implementation approaches. These more continuous forms of engagement are important because they occur throughout the lifecycle of an AI initiative, not only during periodic strategy updates.
OECD countries demonstrate strong engagement with public sector organisations (31 of 36 countries, or 86%) and civil servants (26 of 36, or 72%). For example, France’s social dialogues with central government unions and staff representatives helped to shape commitments for how AI would be introduced in the civil service (Ministry of Transformation and Civil Service, 2024[89]). This is important as they are at the frontline of public-service delivery and their responsibilities are directly impacted by the introduction of AI technologies. Governments also show substantial engagement with broader ecosystem actors, such as academia, industry and civil society (23 of 36, or 64%).
Some countries have more innovative forms of engagement. In 2024, the Belgian Presidency of the Council of the EU convened a representative group of 60 Belgians to collect citizens’ views on AI within the bloc (beEU, 2024[90]). The Belgian government’s AI4Belgium initiative is another promising example of an ongoing, multi-stakeholder ecosystem that brings together a variety of actors to support adoption of AI across sectors and within government (BOSA, n.d.[91]; BOSA, n.d.[92]). A further example, France uses the government’s Agora participation platform to gather large-scale citizen input on national AI priorities, feeding into the work of the country’s AI Commission and its recommendations to public authorities.5
However, there appear to be gaps in two key areas: (1) engaging with service users and (2) cross-border collaboration. Only 16 of 36 OECD countries (42%) engage service users, even though their feedback is crucial to ensure that AI-enabled services are usable and trusted. Only 13 of 36 countries (36%) engage cross‑border actors beyond participation in international fora like the OECD.6 Since date, models and risks are inherently transnational, stronger international co-ordination is needed for interoperability, regulatory alignment and shared learning.
Potential easy opportunities to engage service users are by establishing citizen-complaint mechanisms or other means of gathering feedback. Only 8 of 36 OECD countries (22%) have these. For example, each public sector service provider in Estonia must offer citizens an E-service feedback form, which have now been operationalised as agencies adopt citizen-facing AI solutions.7
While engagement practices appear moderate to solid in most areas, data suggests a disconnect between engagement and action. Significant efforts go into listening to external perspectives, but only half of OECD countries and no accession candidate countries report basing their AI investment decisions on citizen needs or demands.
Meanwhile, engagement efforts continued beyond 2024, such as public consultations in 2025 by Canada (Government of Canada, 2025[93]), Israel (Israel National Digital Agency, 2025[94]) and the United Kingdom (Department for Science, Innovation & Technology, 2026[95]) about AI in government strategies, policies and tools.
Report on the use of AI at central/federal government level to improve public sector functions
Note: 2025 data not available for Germany and the United States. 2023 data not available for Bulgaria, Germany, Greece, Indonesia, Slovak Republic, Switzerland, Thailand and the United States. 2025 data for Indonesia and Thailand cover the period from 1 January 2022 to 31 December 2023. 2023 data not available for the “Oversight and accountability in the PS” category.
Source: OECD (2025), Survey on Digital Government 3.0.
Specific training programmes on skills required to support the development and use of AI in the public sector
Note: Data not available for Germany, Indonesia, Thailand and the United States. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
Availability of key components of digital infrastructure used to support the integration of AI in the public sector
Note: Data not available for Germany, Indonesia, Thailand and the United States. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
Availability of internal controls and mechanisms are in place within the executive branch to ensure accountability in the development and deployment of AI systems by the public sector
Note: Data not available for Germany, the United States, Indonesia and Thailand. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
Public bodies in charge of providing oversight or ethical advice for AI in the public sector, and type of oversight or advice this/these body/bodies provide(s)
Note: Data not available for Germany or the United States. Data cover 1 January 2023 to 31 December 2024. (*) Data for Indonesia and Thailand cover 1 January 2022 to 31 December 2023.
Source: OECD (2025), Survey on Digital Government 3.0.
Availability of central/federal government conducted financial and non-financial impact measurement studies on AI use in the public sector, either prospective, retrospective, or both
Note: Data not available for Germany, Indonesia, Thailand and the United States. Data cover 1 January 2023 to 31 December 2024.
Source: OECD (2025), Survey on Digital Government 3.0.
Actors that collaborated in the process of developing the national strategy, agenda or plan for AI in the public sector
Note: Data not available for Germany and the United States. Data cover 1 January 2023 to 31 December 2024. Data for Indonesia and Thailand cover the period from 1 January 2022 to 31 December 2023.
1. Data were updated following a request from the country after the publication of the 2025 DGI; therefore, these changes are not reflected in the 2025 DGI results.
Source: OECD (2025), Survey on Digital Government 3.0.
References
[11] Agencia Española de Supervisión de la Inteligencia Artificial (AESIA) (n.d.), Ensuring ethical and responsible AI, https://aesia.digital.gob.es/es.
[12] Agencia Española de Supervisión de la Inteligencia Artificial (AESIA) (n.d.), Guides on Artificial Intelligence, https://aesia.digital.gob.es/en/guides.
[13] Agencia Española de Supervisión de la Inteligencia Artificial (AESIA) (n.d.), Organisational structure and governance, https://aesia.digital.gob.es/en/estructura-organizativa.
[27] AI in the State (n.d.), Incubateur ALLiaNCE, https://ia.numerique.gouv.fr/incubateur-alliance/les-produits-incub%C3%A9s/.
[40] ALIA (n.d.), Public AI infrastructure in Spanish and co‑official languages, https://alia.gob.es.
[19] Australian Department of Finance (2025), Introducing the APS AI Plan, https://www.finance.gov.au/about-us/news/2025/introducing-aps-ai-plan.
[85] Australian Government (2024), “Evaluation of the whole-of-government trial of Microsoft 365 Copilot”, https://www.digital.gov.au/initiatives/copilot-trial.
[16] Australian Public Service Academy (n.d.), AI in government fundamentals, https://www.apsacademy.gov.au/course-sessions/ai-government-fundamentals.
[32] Austrian Ministry of Finance (BMF) (n.d.), Predictive Analytics Competence Center, https://www.bmf.gv.at/themen/betrugsbekaempfung/einheiten-betrugsbekaempfung/Predictive-Analytics-Competence-Center.html.
[90] beEU (2024), A Citizens’ View of Artificial Intelligence within the EU, https://glassroots.com/wp-content/uploads/2024/09/Rapport-IA-EN.pdf.
[69] Berryhill, J. et al. (2019), “Hello, World: Artificial intelligence and its use in the public sector”, OECD Working Papers on Public Governance, No. 36, OECD Publishing, Paris, https://doi.org/10.1787/726fd39d-en.
[91] BOSA (n.d.), AI4Belgium, https://bosa.belgium.be/fr/AI4Belgium.
[92] BOSA (n.d.), L’IA dans le secteur public (AI4GOV), https://bosa.belgium.be/fr/AI4Belgium/AI4GOV.
[35] Brazilian Secretariat of Digital Government (2024), Data Maturity Model (MMB), https://www.gov.br/governodigital/pt-br/infraestrutura-nacional-de-dados/maturidade-de-dados/arquivos/modelo-maturidade-de-dados-mmd.pdf.
[71] Bright, J. et al. (2024), Generative AI is already widespread in the public sector, https://arxiv.org/abs/2401.01291.
[9] Brizuela, A. et al. (2025), Analysis of the generative AI landscape in the European public sector, European Commission, https://op.europa.eu/s/z4XY.
[66] Cabinet Office of Japan (2026), Human-Centered AI Society Principles Conference.
[74] Chile’s Ministry of Science, Technology, Knowledge and Innovation (2024), Política Nacional de Inteligencia Artificial, https://drive.google.com/file/d/11OLxLp8NyKgpeRFLe45X0zStY7SFEJIC/edit?pli=1.
[75] Chile’s Ministry of Science, Technology, Knowledge and Innovation (2023), Lineamientos para el uso de Inteligencia Artificial, https://minciencia.gob.cl/uploads/filer_public/ae/9a/ae9a7ce7-807b-4781-9ac3-9b253bfbe735/of_n711_2023_dis_lin_ia_minciencia.pdf.
[54] Congress of the Republic of Peru (2025), Law No. 32270: Amendment to the Organic Law on Elections introducing digital voting.
[67] Consejo Nacional de Política Económica y Social (2025), Política Nacional de Inteligencia Artificial (CONPES 4144), Departamento Nacional de Planeación, https://colaboracion.dnp.gov.co/CDT/Conpes/Econ%C3%B3micos/4144.pdf.
[24] Danish Agency for Digital Government (n.d.), Signature projects with artificial intelligence in municipalities and regions, https://digst.dk/kunstig-intelligens/signaturprojekter.
[25] Danish Agency for Digital Government (n.d.), The Grant Pool for New Technologies, https://digst.dk/digital-transformation/tilskudspuljen-for-nye-teknologier.
[34] Danish Centre for AI Innovation (DCAI) (2024), Denmark’s first AI supercomputer is now operational, https://dcai.dk/news/denmark-s-first-ai-supercomputer-is-now-operational.
[95] Department for Science, Innovation & Technology (2026), Consultation outcome, Guidance for using the AI Management Essentials tool: government response, https://www.gov.uk/government/consultations/ai-management-essentials-tool/outcome/guidance-for-using-the-ai-management-essentials-tool-government-response.
[80] Department of Public Expenditure NDP Delivery and Reform (2024), Interim Guidelines for Use of Artificial Intelligence in the Public Service.
[68] Digital Agency of Japan (2025), Advanced AI Utilization Advisory Board, https://www.digital.go.jp/en/councils/ai-advisory-board.
[52] Digital Agency of Japan (2025), The Guideline for Japanese Governments’ Procurements and Utilizations of Generative AI for the sake of Evolution and Innovation of Public Administration, https://www.digital.go.jp/assets/contents/node/basic_page/field_ref_resources/e2a06143-ed29-4f1d-9c31-0f06fca67afc/6e45a64f/20250527_resources_standard_guidelines_guideline_04.pdf.
[5] Digital Island (n.d.), website, https://island.is/en/o/digital-iceland/island-is.
[57] Digital Transformation Agency of Australia (n.d.), Policy for the responsible use of AI in government, https://www.digital.gov.au/ai/ai-in-government-policy/strategy-and-oversight.
[59] Dutch Government (n.d.), Algoritmeregister (Algorithm Register of the Dutch Government), https://algoritmes.overheid.nl/en.
[3] European Union (2024), Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), https://eur-lex.europa.eu/eli/reg/2024/1689/oj.
[31] Federal Public Service (FPS), Public Social Security Institutions (IPSS) and ICT organisations (n.d.), G-CLOUD, the common cloud of the public sector.
[84] Gartner (2025), Gartner Survey Finds 45% of Organizations With High AI Maturity Keep AI Projects Operational for at Least Three Years, https://www.gartner.com/en/newsroom/press-releases/2025-06-30-gartner-survey-finds-forty-five-percent-of-organizations-with-high-artificial-intelligence-maturity-keep-artificial-intelligence-projects-operational-for-at-least-three-years.
[70] Giesecke, O. (2024), Generative AI Use In US Public Sector On The Rise, Survey By Hoover Fellow Suggests, https://www.hoover.org/generative-ai-use-us-public-sector-rise-survey-hoover-fellow-suggests.
[47] Government of Canada (2026), Algorithmic Impact Assessment tool, https://www.canada.ca/en/government/system/digital-government/digital-government-innovations/responsible-use-ai/algorithmic-impact-assessment.html.
[93] Government of Canada (2025), Consultations on the AI Strategy for the Federal Public Service: What We Heard, https://www.canada.ca/en/government/system/digital-government/digital-government-innovations/responsible-use-ai/consultations-ai-strategy-federal-public-service-what-we-heard.html.
[62] Government of Colombia (2026), Guía Ética para la Implementación, Desarrollo y Uso de Sistemas de Inteligencia Artificial en Entidades Públicas de Colombia, https://www.mintic.gov.co/portal/715/articles-425888_recurso_1.pdf (accessed on 8 June 2026).
[61] Government of Colombia (2025), “Colombia estrena Directiva para garantizar la transparencia de los algoritmos del Estado”, Procuraduría General de la Nación, https://www.procuraduria.gov.co/Pages/colombia-estrena-directiva-garantizar-transparencia-algoritmos-estado.aspx.
[60] Government of Estonia (2026), AI use cases: Examples of artificial intelligent applications in the Estonian public sector.
[23] Government of France (2024), Produits incubés en 2024, https://alliance.numerique.gouv.fr/produit/les-produits-incub%C3%A9s-2024.
[56] Government of France (n.d.), Code des relations entre le public et l’administration, https://www.legifrance.gouv.fr/codes/section_lc/LEGITEXT000031366350/LEGISCTA000031367696/?anchor=LEGIARTI000033205535.
[45] Government of Greece (2024), Blueprint for Greece’s AI Transformation, https://foresight.gov.gr/en/studies/A-Blueprint-for-Greece-s-AI-Transformation/.
[64] Government of Ireland (2025), Guidelines for the Responsible Use of AI in the Public Service, Department of Public Expenditure, Infrastructure, Public Service Reform and Digitalisation, https://assets.gov.ie/static/documents/09fe3ad4/Guidelines_for_the_Responsible_Use_of_AI_in_the_Public_Service_20250918.pdf.
[50] Government of Luxembourg - Ministry of Digitalization (n.d.), The AI4Gov initiative, https://mindigital.gouvernement.lu/en/dossiers.gouv2024_mindigital+fr+dossiers+2021+AI4Gov.html.
[46] Government of Peru (2023), Law No. 31814.
[96] Government of Portugal (2025), Despacho n.º 749/2025, Diário da República, 16 January, https://diariodarepublica.pt/dr/detalhe/despacho/749-2025-903770555.
[97] Government of Portugal (2024), Resolução do Conselho de Ministros n.º 201/2024, Diário da República, 30 December, https://diariodarepublica.pt/dr/analise-juridica/resolucao-conselho-ministros/201-2024-901536075.
[77] Government of the Netherlands (2026), Algoritmekader (Algorithmic Framework), https://minbzk.github.io/Algoritmekader.
[26] Government of the Republic of Türkiye (n.d.), 1711 - Artificial Intelligence Ecosystem Call, Scientific and Technological Research Council of Türkiye – TÜBİTAK, https://tubitak.gov.tr/tr/destekler/sanayi/ulusal-destek-programlari/1711-yapay-zeka-ekosistem-cagrisi.
[86] Government of the United Kingdom (2025), Evaluation of the M365 Copilot pilot in the Department for Business and Trade, Department for Business and Trade, https://www.gov.uk/government/publications/microsoft-365-copilot-pilot-dbt-evaluation-report.
[65] IMY (2025), National guidelines for generative AI in public administration are launched, Swedish Authority for Privacy Protection, https://www.imy.se/en/news/national-guidelines-for-generative-ai-in-public-administration-are-launched/.
[63] Inspector General of Colombia and Ombudsman of Colombia (2025), Estándares de transparencia algorítmica para los sistemas utilizados por el Estado, https://www.defensoria.gov.co/documents/20123/3407303/300925DirectivaConjunta007.pdf/c47e1175-6f60-058a-3e0b-3dfaf82d5f23?t=1759261267112 (accessed on 8 June 2026).
[94] Israel National Digital Agency (2025), Setting the Standard: Israel Unveils First-Ever Responsible AI Guide for the Public Sector, https://www.gov.il/en/pages/ai-guide.
[41] Israel Privacy Protection Authority (2025), Guide on Implementing Privacy-Enhancing Technologies in Artificial Intelligence Systems, https://www.gov.il/BlobFolder/generalpage/pets_ai/en/PETs_AI_english_accessable.pdf.
[17] Korea Internet & Security Agency (KISA) (n.d.), AI Threat Detection and Response, https://academy.kisa.or.kr/cont/programInfo/eduIntroAI.do.
[8] Korean Public Procurement Service (2022), Launch of AI-Based Public Software Project Ordering Support System, https://www.pps.go.kr/kor/bbs/view.do?bbsSn=2208010006&key=00318.
[42] LINC (2025), We tested homomorphic encryption!, https://linc.cnil.fr/teste-le-chiffrement-homomorphe.
[14] Ministry of Industry and Trade, Czech Republic (2024), National Artificial Intelligence Strategy 2030, https://mpo.gov.cz/assets/cz/podnikani/digitalni-ekonomika/umela-inteligence/2024/7/Narodni-strategie-umele-inteligence-2030.pdf.
[20] Ministry of Information Technologies and Communications (MinTIC) (2025), Registration is open for the free diploma in Artificial Intelligence for public servants, https://www.mintic.gov.co/portal/inicio/Sala-de-prensa/Noticias/406073:Abiertas-inscripciones-para-diplomado-gratuito-en-Inteligencia-Artificial-para-servidores-publico.
[22] Ministry of the Economy and Innovation (2025), AI Training Platform, https://eimin.lrv.lt/lt/veiklos-sritys/skaitmenine-politika/dirbtinis-intelektas/di-viesajame-sektoriuje/di-mokymu-platforma.
[82] Ministry of the Interior and Kingdom Relations (2025), Responsible Use of Generative AI: Government-Wide Guide, https://www.government.nl/documents/2025/01/31/responsible-use-of-generative-ai.
[89] Ministry of Transformation and Civil Service (2024), Towards a strategy to use Artificial Intelligence (AI) in French civil service HRM.
[73] MIT NANDA (2025), The GenAI Divide: State of AI in Business 2025, MIT Networked AI Agents in Decentralized Architecture, https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf.
[78] New Zealand Digital Government (2025), Responsible AI guidance for the Public Service: Generative AI – Customer experience and Privacy, https://www.digital.govt.nz/standards-and-guidance/technology-and-architecture/artificial-intelligence/responsible-ai-guidance-for-the-public-service-genai/customer-experience/privacy.
[79] New Zealand Government (n.d.), Data standards toolkit, https://data.govt.nz/toolkit/data-standards.
[43] NHS Digital (n.d.), Artificial data pilot, https://digital.nhs.uk/services/artificial-data.
[72] OECD (2026), Building an AI-ready public workforce: Implications and strategies, OECD Publishing, Paris, https://doi.org/10.1787/b89244c7-en.
[10] OECD (2025), “AI openness: A primer for policymakers”, OECD Artificial Intelligence Papers, No. 44, OECD Publishing, Paris, https://doi.org/10.1787/02f73362-en.
[4] OECD (2025), Digital Government Review of Korea: Harnessing Digital and Data to Transform Government, OECD Digital Government Studies, OECD Publishing, Paris, https://doi.org/10.1787/9defc197-en.
[2] OECD (2025), Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, OECD Publishing, Paris, https://doi.org/10.1787/795de142-en.
[30] OECD (2025), “Sharing trustworthy AI models with privacy-enhancing technologies”, OECD Artificial Intelligence Papers, No. 38, OECD Publishing, Paris, https://doi.org/10.1787/a266160b-en.
[28] OECD (2024), “AI, data governance and privacy: Synergies and areas of international co-operation”, OECD Artificial Intelligence Papers, No. 22, OECD Publishing, Paris, https://doi.org/10.1787/2476b1a4-en.
[88] OECD (2024), “Framework for Anticipatory Governance of Emerging Technologies”, OECD Science, Technology and Industry Policy Papers, No. 165, OECD Publishing, Paris, https://doi.org/10.1787/0248ead5-en.
[1] OECD (2024), “Recommendation of the Council on Artificial Intelligence”, OECD Legal Instruments, OECD/LEGAL/0449, OECD, Paris, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449.
[29] OECD (2023), “Emerging privacy-enhancing technologies: Current regulatory and policy approaches”, OECD Digital Economy Papers, No. 351, OECD Publishing, Paris, https://doi.org/10.1787/bf121be4-en.
[55] OECD (2023), Global Trends in Government Innovation 2023, OECD Public Governance Reviews, OECD Publishing, Paris, https://doi.org/10.1787/0655b570-en.
[87] OECD (forthcoming), Governing with Artificial Intelligence: Measuring impact and demonstrating returns on AI investments.
[7] OECD Observatory of Public Sector Innovation (OPSI) (n.d.), Robot Alice – Bid, Contract and Notice Analyser, https://oecd-opsi.org/innovations/robot-alice-bid-contract-and-notice-analyser/ (accessed on 1 June 2026).
[38] Office for National Statistics (ONS) - Data Science Campus (2023), Synthesising the linked 2011 Census and deaths dataset while preserving its confidentiality, https://datasciencecampus.ons.gov.uk/synthesising-the-linked-2011-census-and-deaths-dataset-while-preserving-its-confidentiality/.
[83] Ramos, L. and R. Kandaswamy (2023), Capture AI Value With These 5 Benefit Realization Best Practices, Gartner, https://static1.squarespace.com/static/5530dddfe4b0679504639dc1/t/66040dae666c427d843c6de6/1711541693913/Capture+AI+Value+With+These+5+Benefit+Realization+Best+-+Gartner.pdf.
[33] Republic Portuguesa (2023), Virtual assistant with Artificial Intelligence will support citizens in ePortugal, https://www2.gov.pt/en/noticias/assistente-virtual-com-inteligencia-artificial-vai-apoiar-os-cidadaos-no-eportugal.
[76] Secretariat of Digital Government of Brazil (2025), Cartilha de Inteligência Artificial Generativa (Generative AI Guide), https://www.gov.br/governodigital/pt-br/infraestrutura-nacional-de-dados/inteligencia-artificial-1/publicacoes/cartilha-ia-generativa.
[81] Spanish Agency for the Supervision of Artificial Intelligence (AESIA) (n.d.), Ensuring ethical and responsible AI, https://aesia.digital.gob.es.
[36] Sweden Agency for Digital Government (DIGG) (n.d.), Ena – Sweden’s digital infrastructure, https://www.digg.se/styrning-och-samordning/ena---sveriges-digitala-infrastruktur.
[51] TSE Global (n.d.), Sistem Belgelendirme (System Certification Services), https://www.tseglobal.com.tr/sistem-belgelendirme?num=38%3Flg%3Den.
[49] Turkish Standards Institution (TSE) (2026), Ayna Komiteler (Mirror Technical Committees), https://intweb.tse.org.tr/standard/aynakomite/aynakomiteler.aspx.
[48] Türkiye Digital Transformation Office and Ministry of Industry and Technology (2021), National Artificial Intelligence Strategy 2021–2025, https://wp.oecd.ai/app/uploads/2021/12/Turkey_National_Artificial_Intelligence_Strategy_2021-2025.pdf.
[18] UK Cabinet Office (2025), One Big Thing 2025: ‘AI for All’ - A Modern Civil Service Blog,, https://moderncivilservice.blog.gov.uk/2025/10/14/one-big-thing-ai-for-all-is-live/.
[44] UK Department for Science, Innovation and Technology (2026), AI Opportunities Action Plan: One Year On, https://www.gov.uk/government/publications/ai-opportunities-action-plan-one-year-on/ai-opportunities-action-plan-one-year-on.
[39] UK Department of Science, Innovation and Technology (2023), The UK-US Blog Series on Privacy-Preserving Federated Learning: Introduction, https://rtau.blog.gov.uk/2023/12/07/the-uk-us-blog-series-on-privacy-preserving-federated-learning-introduction/.
[58] UK Government (n.d.), Find out how algorithmic tools are used in public organisations, https://www.gov.uk/algorithmic-transparency-records.
[6] UK Government (n.d.), Government – AI Incubator for Artificial Intelligence (i.AI), https://ai.gov.uk/our-work/government/ (accessed on 1 June 2026).
[37] UK Government (n.d.), Service Standard, https://www.gov.uk/service-manual/service-standard.
[53] UK Government Digital Service (n.d.), Making the Algorithmic Transparency Recording Standard (ATRS) mandatory across government, https://dataingovernment.blog.gov.uk/2025/05/08/making-the-algorithmic-transparency-recording-standard-atrs-mandatory-across-government.
[21] UNESCO (2025), Strengthening Digital Resilience: Costa Rica’s AI Push After a Massive Cyberattack, https://www.unesco.org/es/articles/fortalecer-la-resiliencia-digital-el-impulso-de-costa-rica-por-la-ia-tras-un-ciberataque-masivo.
[15] University of Helsinki and MinnaLearn (n.d.), Elements of AI – Free online course on artificial intelligence, https://www.elementsofai.com.
← 1. For more information on AI in Government, see https://oecd.ai/gov.
← 2. The OECD.AI Policy Navigator is available at https://oecd.ai/dashboards. More specifically, AI use cases in the public sector can be found at https://oecd.ai/dashboards/policy-initiatives?initiativeTypeIds=123.
← 3. For more information, see https://www.oecd.org/en/topics/sub-issues/privacy-enhancing-technologies.html.
← 4. Portugal’s National Agenda on Artificial Intelligence, as part of the National Digital Strategy: Resolution 749/2025 (Government of Portugal, 2025[96]) and Legal Analysis of Resolution 201/2024 (Government of Portugal, 2024[97]).
← 5. Acteurs Publics (2026), IA : David Amiel veut faire aboutir le dialogue social à un accord d’ici l’automne (AI: David Amiel aims to reach a social dialogue agreement by autumn).
← 6. The rates among OECD accession candidate countries are: public sector organisations (50%); civil servants (17%); broader ecosystem (33%); service users (17%); and cross-border actors (17%).
← 7. See Estonia’s guidelines for digital public services (https://digiriik.eesti.ee/protsess/kasutuselevott/tagasiside-kogumine) and the Estonian Tax and Customs Board’s online feedback form (https://www.emta.ee/en/private-client/board-news-and-contacts/contacts/feedback).