Artificial intelligence (AI) is increasingly embedded in how governments design policies, deliver services and manage public administration. As public institutions move to actively procuring, developing and using AI public perceptions of these technologies become a critical factor shaping their legitimacy and effectiveness. This chapter examines how people across OECD member countries view government use of AI and what this means for trust in public institutions, drawing on evidence from the 2025 OECD Survey on Drivers of Trust in Public Institutions.
OECD Survey on Drivers of Trust in Public Institutions 2026 Results
Navigating Rising Expectations and New Horizons
Abstract
Most people remain sceptical of AI deployment in the public sector, although the high rate of neutral or ‘do not know responses suggest lack of information or experience preventing the ability to form a judgement.
People are more likely to believe that AI in the public sector can improve service quality and efficiency than to believe it will be fair, transparent, and protective of personal information. Optimism about government use of AI aligns more closely with trust in public institutions at the individual level than at the country level. This means that people in high trust countries (i.e. countries in which a comparatively elevated share of people have high or moderately high trust in public institutions) are not necessarily more positive about AI, whereas high trust individuals (i.e. people who report high to moderately high trust in the government) tend to be.
Greater knowledge of AI is associated with higher optimism about government’s use of it.
The OECD Trust Survey results support the public governance levers set out in the OECD Framework for Trustworthy AI in Government, which helps governments align their use and governance of AI with the OECD AI Principles. Confidence in data handling by public bodies, stewardship of long-term issues, and appropriate regulation of new technologies are associated with a more optimistic view of government use of AI.
Artificial intelligence (AI) is increasingly transforming public sector activity, reshaping processes and outcomes across a wide range of government functions. In this chapter, AI refers to computer systems that can carry out tasks typically associated with human intelligence, including understanding language, learning from data, and supporting decision making. More formally, the OECD defines an AI system as a
“a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment.” [OECD/LEGAL/0449].
In OECD Member countries, thousands of AI-related projects are now underway, intensifying both public and political attention (OECD, 2025[1]). In addition to their traditional role as regulators in AI, public administrations are now procurers, operational users, and developers of AI systems. They are integrating AI into policy design, public service delivery, regulatory oversight, and back-office processes as part of efficiency-enhancing government modernisations.
As governments deploy AI, the relationship between the state and the public may shift. AI systems increasingly mediate day-to-day interactions with individuals and shape how governments respond to complex policy challenges. Used well, these technologies offer real opportunities to improve productivity, responsiveness and accountability. However, if not developed and used in a trustworthy manner, public institutions face risks that could result in serious harms for individuals and society, potentially undermining the legitimacy of government institution’s use of AI and public trust. The costs of inaction are also growing. Failing to engage with AI may limit democratic governments' capacity to deliver public value in a rapidly evolving technological landscape, widening the gap between public and private sector capabilities (OECD, 2025[1]).
The recent OECD report Governing with Artificial Intelligence, and the OECD AI principles which underpin it, underscores that “trustworthiness” is a precondition for the effective and responsible use of AI in the public sector. It identifies public resistance as a potential challenge in almost half of use cases reviewed, reflecting concerns sometimes shaped in part by previous high-profile failures in algorithmic decision making. Addressing this challenge, and the underlying concerns it signals, requires robust governance frameworks that provide clear accountability and avenues for redress, as well as continuous monitoring, oversight, and strong risk-management processes (OECD, 2025[1]). The report sets out three complementary pillars - enablers, guardrails, and engagement - which together form the OECD Framework for Trustworthy AI in Government. This framework seeks to align public-sector AI development and deployment with OECD AI Principles/Recommendation of the Council on Artificial intelligence [OECD/LEGAL/0449] (adopted in 2019, updated in 2024).
Insights from the OECD Trust Survey provide an empirical basis for understanding how public expectations align – or diverge – from current practices. By grounding AI strategies and related activities in evidence about how they are perceived by the public, administrations can better anticipate and mitigate risks and challenges related to public resistance, or inaction. Doing so helps safeguard the potential benefits of AI adoption and supports more trusted, human-centred implementation.
This chapter examines public perceptions of government use of AI and its implications for public trust in public institutions. It begins by presenting overall attitudes towards the use of AI in the public sector, and highlighting cross‑country patterns of optimism, scepticism, and uncertainty. The chapter then compares these public expectations with the current landscape of government AI deployment, drawing on findings from Governing with Artificial Intelligence. It next explores how socio‑economic characteristics, lived experience, and familiarity with AI shape attitudes toward its adoption in public services. Building on these insights, the chapter assesses key public governance levers that are associated with more positive expectations of government AI use. Together, these sections show that while AI has the potential to enhance trust in public institutions, this outcome depends critically on how governments deploy, regulate, and communicate about AI systems.
To assess public trust in governments’ use of AI, public administrations can track public attitudes toward AI over time to identify where government practice and public expectations may diverge. To this end, the 2025 OECD Trust Survey introduced a new survey question designed to measure public attitudes towards government agencies’ use of artificial intelligence. Respondents were asked to assess the likelihood that government agencies will fulfill six different expectations when using AI:
provide services better tailored to individual needs,
reduce costs,
protect individuals’ personal information from unauthorised access or misuse,
maintain human oversight and final decision-making authority in critical areas,
be transparent about how and where AI is used in government operations,
and ensure fair and unbiased treatment of the population.
For ease of presentation, responses are grouped into four respondent profiles based on how many expectations respondents consider likely to be met when government agencies use AI. A very positive assessment refers to respondents who believe five or six of the above expectations will be met. A moderately positive assessment refers to those who believe three or four expectations will be met. A moderately skeptical assessment refers to those who believe only one or two expectations will be met. A very skeptical assessment refers to respondents who believe none of the expectations will be met. These labels are used throughout the chapter as a shorthand to describe patterns in attitudes towards government use of AI.
OECD Trust Survey results reveal scepticism across the OECD and the European OECD accession candidate countries regarding the potential benefits of AI deployment in government. Selecting zero expected positive expectations of AI use by government, the response for 35% of respondents, is by far the most common result, on average, across OECD countries (Figure 5.1). In contrast, 22% of people have a very positive assessment, 19% a moderately positive assessment, and 25% a moderately skeptical assessment. The shares in the different groups are very similar across the European OECD accession countries1, with 22% having a very positive, 19% a moderately positive, 23% a moderately skeptical and 37% having a very skeptical assessment.
The survey also reveals significant variation by country. Over 40% of people in Estonia, Iceland, Ireland, Latvia, Norway, the Slovak Republic and the United Kingdom, as well as in Bulgaria, express low confidence in their governments’ (potential) use of AI. By contrast, Korea and Switzerland have notably higher shares of respondents expressing positive expectations. These geographic differences suggest that a variety of different factors may shape public confidence in government use of AI, including different experiences with digital government services and distinct cultural attitudes towards technological innovation in the public sector. At the country level (in contrast to the individual level), more prevalent scepticism towards government AI use is not necessarily related to low levels of trust in different public institutions. In other words, high-trust countries do not always have higher rates of optimism towards government use of AI, and lower-trust countries do not always express a more skeptical outlook.
Note: This figure presents confidence in how government agencies will implement and use artificial intelligence. It shows the share of respondents who are confident (responses 6-10 on a 0-10 scale) that government use of AI will fulfil a different number of positive expectations, in response to the question: “Thinking about how government agencies could use AI in the future, how confident are you that they will achieve the following outcomes?”. Panel A presents the OECD average distribution of respondents across scores from 0 to 6, where each score reflects the number of expectations in which respondents are confident. Panel B shows country-level shares grouped into four categories: 5-6 expectations (very positive), 3-4 (moderately positive), 1-2 (moderately skeptical), and 0 expectations (very skeptical); the OECD value represents the average of these country shares. The six expectations are: providing services better tailored to individual needs, reducing costs, protecting personal information from unauthorised access or misuse, maintaining human oversight and final decision-making authority in critical areas, ensuring transparency in AI use, and guaranteeing fair and unbiased treatment of the population.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
Efforts to gather evidence on public perceptions of government use of AI at the national level are still in their infancy, but some countries have begun to collect such data (see Box 5.1). This type of evidence supports the engagement pillar of the OECD framework for trustworthy AI in government by helping public administrations understand public expectations and tailor participation and communication on how AI is used in government.
Australia’s long-term insights briefing “How might artificial intelligence affect the trustworthiness of public service delivery”
Australia’s long-term insights briefing “How might artificial intelligence affect the trustworthiness of public service delivery?” was published in 2023 and examines how the Australian Public Service (APS) could integrate AI into future service delivery and what that could mean for the perceived trustworthiness of those services.
Methodologically, it draws on multiple evidence streams: community engagement and expert input, including workshops and focus groups with participants from community organisations, academia, industry, youth and APS agencies; a scenario-based workshop facilitated by the Australian National University’s National Security College Futures Hub to test plausible future scenarios and their implications for trustworthiness; two surveys that together gathered over 5 000 responses; and a Rapid Evidence Assessment of existing literature and research.
United Kingdom Public Attitudes to Data and AI tracker survey
In the United Kingdom, the Centre for Data Ethics and Innovation has been engaging the public through the Public Attitudes to Data and AI tracker survey, which can inform the government’s approach to future policy development. Four waves of the survey have been carried out.
Belgium panel on AI
In 2024, the Belgian Presidency of the Council of the European Union convened 60 randomly selected Belgian citizens on a panel to discuss the future of artificial intelligence (AI) in the EU. The outcome of their discussion was summarised in a report outlining citizens’ view of AI within the EU.
Source: adapted from (OECD, 2025[1]; Government of Australia, 2023[2]).
Beyond this general snapshot of the population’s attitudes towards AI in the public sector, a distinct pattern emerges: people express more confidence that government use of AI will improve responsiveness and efficiency (more tailored services, lower costs) than in its ability to uphold procedural values such as transparency, privacy or fairness (Figure 5.2).
Overall confidence levels are highest for service improvement (43%) and cost reduction (42%), suggesting that a substantial part of the public broadly recognises AI’s potential efficiency and service improvement benefits.
Confidence drops when it comes to value-based concerns over processes such as protecting personal information (32%), ensuring transparency (34%) and guaranteeing fair treatment (35%). People’s views on whether they believe that humans would maintain oversight of critical decisions are slightly more balanced, with 37% finding it likely and 36% unlikely that they would.
These patterns are broadly consistent across the different OECD and European OECD accession candidate countries.
This aligns closely with findings from the 2024 Risks That Matter Survey (OECD, 2025[3]). For example, 40% consider the use of AI to help process and approve social programme applications to be good for users, while 30% report uncertainty and 25% believe it is not good for users. Attitudes toward data governance are similarly mixed: 37% disagree or strongly disagree that they trust the government with data collected about them through digital tools and AI, compared with 32% who indicate that they do. Finally, fewer than half of respondents (44%) were confident that AI would be used to assess public benefits applications only when it is safe and trustworthy to do so.
In the OECD Trust Survey, equally notable is the high proportion of neutral and “do not know” responses across all dimensions. On average across the OECD, but with very similar findings for the European OECD accession candidate countries, these range from 9 to 11% for “do not know” alone and reach 25% when combined with neutral responses. This uncertainty suggests a significant portion of the public remains either uninformed about how governments are or might be deploying AI systems, or unable to form confident judgements about these current and potential future applications.
Share who are or are not confident in the potential of government AI use fulfil the respective expectation, OECD, 2025
Note: The figure shows the unweighted OECD averages of weighted OECD country averages of responses to the question “Thinking about how government agencies could use AI in the future, how confident are you that government agencies will achieve the following outcomes. “Confident” corresponds to a response of 6-10 on the 0-10 scale, “neutral” to 5 and “not confident” to 0-4 on the 0-10 scale. The OECD average is calculated as the unweighted average of weighted country averages. The sub-questions are: providing services better tailored to individual needs, reducing costs, protecting personal information from unauthorised access or misuse, maintaining human oversight and final decision-making authority in critical areas, ensuring transparency in AI use, and guaranteeing fair and unbiased treatment of the population.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
Taken together, the evidence points to a multiplicity of expectations and comfort levels around AI-enabled services and policies. A sizeable share of people feel unsure about AI and have reservations about the outcomes of its deployment by their government. This underlines the importance of a measured, human-centered approach: offering clear explanations, building confidence over time, and maintaining a multi-channel approach to ensure no one is left out.
To better understand how AI might affect public trust, it is helpful to compare what the public expects with the way in which government agencies use AI in practice. To do so, this section draws on use cases described in the Governing with Artificial Intelligence report. This approach enables us to assess the extent to which public attitudes and expectations seem to align with current patterns of government AI use. The Governing with AI report uses case studies from countries that participated in the OECD Trust Survey and from some that did not2.
Out of 200 AI use cases, the Governing with Artificial Intelligence report found that the four primary domains where governments have concentrated their AI efforts were public service design and delivery, civic participation, justice administration, and law enforcement and disaster risk management (see Figure 5.3). Governments also typically apply AI across four distinct types of activities: public-facing service delivery, internal operations, oversight (both internal and external), and policymaking support. Public-facing service delivery emerged as the dominant use case, followed closely by internal operations (OECD, 2025[1]).
Source: OECD (2025[1]), 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; and OECD analysis of identified use cases.
As described in the previous section of this chapter, people express greater confidence in AI's ability to lead to better tailored services at a lesser cost; 43% expect service improvement and 42% expect cost reductions for government. The larger share of use cases deployed in public service design and delivery, and in civic participation suggest governments have in part concentrated their efforts in areas of relatively strong public acceptance. These functions of government create a lower-risk entry point for administrations: by demonstrating clear, tangible value in domains that the public already supports, governments can build credibility, strengthen trust, and lay the groundwork for expanding AI use into more sensitive or complex areas over time.
Taken together, this evidence points to clear pathways through which AI use in government could positively affect trust in public institutions. Performance improvements in public services are one of the most direct channels: public-facing service delivery is the predominant application area, and 45 of 200 documented use cases focus on public service design and delivery. This matters as satisfaction with administrative services is linked to higher trust in all four institutions, with the largest associations for trust in the civil service (5.4 percentage points) and local government (4.1 percentage points).
Emerging use cases help illustrate what these improvements can look like in practice (OECD, 2025[1]). For example, France’s Services Publics+ platform points to a complementary pathway, using AI (e.g., speech-to-text, drafting support, and feedback analysis) to improve accessibility and responsiveness while keeping public servants in control of final decisions (Government of France, 2026[4]).
AI-facilitated civic participation, the second most common type of government AI use case observed (see Figure 5.3), could lead to more effective participatory processes, which in turn could raise levels of public trust. In OECD countries, only 31% of respondents feel that the political system lets people like them have a say in what government does, and 32% feel governments are likely to adopt the proposals emanating from public consultations (see Chapter 6). AI deployed for civic participation could help address this by synthesising insights from large volumes of public consultation input, generating responses and updates to participants, enabling faster fact-checking, detecting misinformation, and making complex information more accessible to the public (OECD, 2025[1]; Adam, 2023[5]). All of these processes can help governments respond more quickly and accurately to public preferences and demands.
Regression analysis of OECD Trust Survey results suggest that these innovations could increase trust in public institutions (see Chapter 1). Government adoption of public opinion in consultations is associated with higher trust in national government (2.0 percentage points), while broader perceptions of political efficacy show strong associations with trust in the national government and legislature (3.2-3.4 percentage points), and weaker ones with trust in local government (1.7 percentage points). If AI tools can make participation feel more influential, this could translate into wider gains in institutional trust.
AI deployment in justice administration and law enforcement are the third and fourth most common government functions for AI applications (Figure 5.3). AI can notably help justice systems by streamlining administration, providing data-driven insights, supporting legal inquiry, easing backlogs and complex procedures while improving public access which represent a high value added sector for the wider use of AI. However, these government functions are sensitive and typically subject to more public resistance.
Nevertheless, to achieve these positive outcomes, the Governing with AI report names a number of preconditions to mitigate the risks associated with the use of AI (Box 5.3). Among them are ensuring that the data on which AI models are built are sufficiently complete, that the algorithms used are transparent and that decisions remain explainable and contestable (OECD, 2025[1]). Spain and the United Kingdom have developed tailored approaches to address these issues (see Box 5.2).
Two countries have taken distinct approaches to governing AI use in their judicial systems. The United Kingdom has issued guidance for judicial office holders, while Spain has established a comprehensive and binding national policy. Both frameworks aim to balance innovation with judicial integrity while addressing risks such as bias, confidentiality, and accountability.
United Kingdom
The UK AI Judicial Guidance provides direction for judicial office holders on appropriate AI use in courts and tribunals. It applies to all judicial personnel under the Lady Chief Justice and Senior President of Tribunals. The guidance provides principles, identifies acceptable uses like text summarization and administrative tasks, while cautioning against reliance on AI for legal research. It emphasises confidentiality, accountability, bias mitigation, and security, and warns against AI-generated misinformation in proceedings.
Spain
In June 2024, Spain approved its National Policy on the Use of Artificial Intelligence in the Administration of Justice. Grounded in the Justice Digital Efficiency Act, the policy establishes a framework for responsible AI use across the entire judicial process. It draws a clear line between systems that could affect judicial independence and those intended purely for administrative purposes. It permits AI for document summarisation and administrative automation but prohibits automated decision-making without human oversight and the generation of binding legal content from protected data. Oversight is divided between the General Council of the Judiciary (CGPJ) for judicial functions and CTEAJE for administrative contexts, with mandatory algorithmic audits and FAT (Fairness, Accuracy, and Transparency) registers.
Source: adapted from OECD (2025[1]), 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.
As seen in this section, the current distribution of AI use in the public sector aligns closely with areas where citizens express the greatest optimism about AI's potential benefits, such as more tailored services and greater efficiency. This suggests that governments’ AI efforts may contribute to greater public acceptance. Moreover, mapping deployment patterns against Trust Survey results indicates that public sector use of AI has the potential to improve governance performance and public perception of key drivers of trust in public institutions.
At the same time, this alignment is neither automatic nor guaranteed. The OECD report Governing with AI underscores that if public authorities manage AI risks poorly, the potential for error and backlash is significant and could undermine trust (OECD, 2025[1]). The box below summarises the main risks associated with AI, and the specific ways these risks can be heightened in government use cases.
The global adoption of AI across sectors raises questions about trust, fairness, privacy, safety, and accountability, among others, which may necessitate specific governance processes or mitigation measures. It is therefore important for governments to identify and manage these risks, consider how AI systems may impact people, help ensure the delivery of AI benefits and mitigate potential harm. The Governing with AI OECD report identifies five general types of risks for the use of AI in government, distinct from any implementation challenges:
Ethical risks: These include AI uses that undermine the free exercise of human rights and freedoms, including privacy, potentially infringing on human-centred values either deliberately or inadvertently. AI algorithms can introduce ethical risks from the digital realm to the physical world through biased algorithms and unethical behaviours like invasive surveillance. Key concerns include threats to trust, fairness, freedom, dignity, individual autonomy and labour rights.
Operational risks: These include technical and operational failures that might affect data privacy, the quality of AI outcomes and internal government operations due to cybers threats, unintended consequences, hallucinations, systematic errors and overreliance on AI systems.
Exclusion risks: These risks relate to gaps that arise when citizens without access to technology or digital literacy can be left behind and unable to benefit from AI advancements in public services.
Public resistance risks: These include public resistance to government use of AI. This can be driven by distrust in government AI systems or processes, or by the spread of false or misleading information about how AI is implemented in public administrations and its potential impacts.
Risks of inaction: Although often overlooked, this risk includes government delays in using AI to yield positive benefits. This can result in significant financial and non-financial costs — which could have otherwise been avoided with AI’s successful adoption — and a widening gap between public sector and private sector capabilities.
Source: adapted from OECD (2025, pp. 32-35[1]), 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.
These concerns are not merely hypothetical. The OECD AI Incidents Monitor data shows a steady increase in incidents and hazards covered by media outlets, and in January 2026, one in five recorded incidents (115 out of 599 incidents) relates to “government, security and defence” underscoring the very real risks associated with public sector deployment of AI (OECD, 2026[6]). One example of real-world harm is the Netherlands’ Toeslagenaffaire in which an AI system used by the government led to 26 000 families being wrongfully accused of fraud due to an algorithm that disproportionately targeted families with dual nationalities or migrant backgrounds (OECD, 2025[1]).
Of course, many of the risks associated with AI are not entirely new. In practice, AI systems often amplify familiar governance failures, such as over-reliance on technical tools, weak accountability, and the uncritical acceptance of outputs as authoritative. The British Post Office scandal, in which hundreds of sub-postmasters being wrongly prosecuted after faults in an accounting system generated incorrect shortfall data for example, was not AI-related, but shows how serious harm can result from unchecked software-generated information (Growns et al., 2024[7]).
Nevertheless, it is also noteworthy that the government functions in which AI use cases are the most common (service design and delivery) are also the areas in which people see the highest potential. As AI applications by government institutions become more common, the share of the population with positive assessments may rise. It may also not, as greater awareness of AI use could heighten perceptions of risk, potentially unevenly across the population.
5.3.1. Patterns of optimism towards government use of AI mirror several of the socio-economic and demographic gaps observed in trust in the national government
In the context of AI-enabled reforms or services, uneven confidence in the public sector's use of AI could trigger resistance among those with the most sceptical views or widen perceptions of government competence and values across a country’s population, with implications for trust.
OECD Trust Survey data show consistent demographic and socio-economic patterns in attitudes toward public sector use of AI across OECD countries (see Figure 5.4). Scepticism increases with age, rising from 24% among 18–29-year-olds to 41% among those 50 and older. Positive attitudes on the other hand remain relatively flat across age groups. Scepticism drops from 42% among those with lower levels of formal education to 28% among the highly educated, while positive attitudes rise from 18% to 27%. Women express higher scepticism (38% vs. 31% for men) and lower positive sentiment (18% vs. 25% for men). Lastly, individuals who feel they may face discrimination or feel financially insecure show higher rates of scepticism (40% and 37% respectively) and lower positive attitudes (20 and 19%).
When comparing the magnitude of these gaps, education emerges as the strongest differentiator, with a 14 percentage point spread in scepticism and 9 percentage points in positive attitudes. Financial concerns produce the same size gap as education with regards to positive attitudes (9 percentage point gap), while age generates the widest scepticism divide at 17 points, despite minimal difference in positive sentiment. This pattern suggests structural factors such as education and economic security shape attitudes towards AI more strongly than demographic characteristics do alone.
These observed patterns of optimism towards government use of AI mirrors several of the socio-economic and demographic gaps observed in trust in the national government. In particular, the same groups that report lower trust, those experiencing financial insecurity and those who feel they may face discrimination, also tend to be more sceptical and less positive about government use of AI. Education shows a similarly consistent pattern: higher educational attainment is associated with both higher trust in national government and more favourable expectations about AI in government (lower scepticism and higher positive attitudes). Taken together, these overlaps suggest that attitudes towards government AI are not only about the technology itself but also reflect broader lived experiences with institutions.
At the same time, the comparison also points to important differences, notably by age and (to some extent) between men and women. Trust in national government varies only modestly by age, and is slightly higher among older respondents, whereas scepticism about AI rises sharply with age. This divergence implies that age-based gaps in AI attitudes may be driven less by general institutional trust and more by AI-specific concerns such as unfamiliarity, perceived loss of human oversight over decision making, or heightened sensitivity to potential harms. Women also appear more sceptical and less positive about government AI than men, and this gap may be larger than the corresponding trust gap, suggesting that AI may amplify concerns that are not fully captured by general trust measures.
Finally, it is worth noting that the previously explained finding that people are more optimistic about AI’s potential responsiveness and efficiency benefits in government (more tailored services and lower costs) than about value-based aspects (privacy, transparency, and fairness) holds true across different demographic groups.
Note: The figure shows the OECD average for the share who have a very positive (panel A) or very skeptical (panel B) perspective on the potential for government agencies’ use of artificial intelligence to fulfil positive expectations among the respective population group. The OECD average is calculated as the unweighted average of weighted country averages. A very positive assessment is defined as being confident in five to six items, with confidence corresponding to a 6-10 response on the 0-10 scale. A very skeptical assessment is defined as being confident with regards to zero items. The underlying question is “Thinking about how government agencies could use AI in the future, how confident are you that government agencies will achieve the following outcomes?”. The items are providing services better tailored to individual needs, reducing costs, protecting individuals’ personal information from unauthorised access or misuse, maintaining human oversight and final decision-making authority in critical areas, being transparent about when and how AI is used, and ensuring fair and unbiased treatment of the population.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
5.3.2. Greater familiarity with AI is associated with higher expectations of its benefits
Despite its apparent pervasiveness in every facet of our lives, most people have only a general or no understanding of what AI is, which poses a significant challenge for public bodies deploying AI technologies. Even with comprehensive frameworks for trustworthy AI in place, low familiarity with AI risks increasing perceptions of opacity in government affairs, highlighting the tension between the rapid pace of AI implementation and the slower pace of public understanding and democratic processes. Moreover, these findings underscore the importance of the "engagement" pillar of the OECD Framework for Trustworthy AI in Government, which emphasises user-centred and responsive approaches. By actively engaging stakeholders, including the public, civil society, and businesses, governments can enhance public understanding of AI and its implications for governance (OECD, 2025[1]).
Individual interests and experiences with AI are increasingly seen as potential predictors of AI attitudes (Kanzola, Papaioannou and Petrakis, 2024[8]). On average across the OECD, fewer than one in ten (8%) report not being familiar with AI, while four in ten (40%) report understanding it well enough to explain it to others and half (50%) say they have a general idea of what it means (see Figure 5.5). These patterns are very similar in the European OECD accession candidate countries. Familiarity tends to be higher among younger people and those with higher levels of education, while older and less-educated groups more commonly report limited understanding or no familiarity. In contrast, differences in awareness or understanding between men and women or people with and without financial concerns are present but more modest.
Share of the population by self-reported knowledge of artificial intelligence, OECD, 2025
Note: The figure shows the within-country distribution of answers to the question “How familiar are you with the term “artificial intelligence (AI)”?. “OECD” is the unweighted average of the weighted OECD country averages.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
Familiarity with artificial intelligence is associated with higher expectations about the potential benefits of AI in government, although even well-informed users have positive expectations of government use of AI in only two areas (see Figure 5.6). Among people who feel they understand AI well enough to explain it to others, three in ten are very optimistic (having positive expectations in five to six areas), compared with around two in ten among those who only have a general idea of what the term means. By contrast, people who are least familiar with AI - and whose only knowledge may stem from the explanation3 provided in the survey following the question on how well they know AI - tend to express much lower confidence, with only around one in ten holding such views. This pattern aligns with the existing literature, which finds that people with stronger AI-related interests or experience often evaluate AI more positively (Kim and Lee, 2023[9]; Kovačević and Demić, 2023[10]).
Regardless of whether respondents report being somewhat or very familiar with the term ‘artificial intelligence’, they are more likely to expect AI to improve services and reduce costs than to expect it to be fair and transparent, protect personal information, and remain subject to human oversight. Among people who are unfamiliar with the term, the assessments are much more uniformly negative (see Figure 5.6).
Share of respondents who are confident that AI use by government agencies can fulfil positive expectations for the respective item, by their self-assessed knowledge of AI, OECD average, 2025
Note: The figure shows the share of “confident” responses to the question “Thinking about how government agencies could use AI in the future, how confident are you that government agencies will achieve the following outcomes. “Confident” corresponds to a response of 6-10 on the 0-10 scale. The OECD average is calculated as the unweighted average of weighted country averages.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
Taken together, these findings suggest caution when interpreting results on “optimism” or “scepticism” toward government use of AI as fully informed, stable attitudes: for many respondents, “AI” may function as a broad shorthand shaped by partial information, media narratives, or general trust in government and technology. Low familiarity can also amplify perceptions of opacity in government decision-making, meaning that expressed scepticism may reflect concerns about transparency, control, and accountability as much as views on any specific AI application. At the same time, these perceptions, even when imperfect or based on limited understanding, still shape people’s lived experiences of public services and, ultimately, their trust in public institutions. This is precisely the value of a perception survey: it captures how government action is received and interpreted by people, complementing objective performance indicators that measure what systems do in practice.
Overall, these findings highlight that public attitudes toward AI in government are shaped by structural factors such as education, economic security, and familiarity with digital technologies. Uneven understanding and experiences with AI risk reinforcing existing divides if governments do not proactively address concerns around fairness, transparency, and accountability. Building acceptance therefore requires not only trustworthy systems, but also meaningful engagement with the groups most likely to feel skeptical.
Socioeconomic factors play a role in perceptions of government use of AI, yet government actions can strongly influence public acceptability. The OECD Framework for Trustworthy AI highlights concrete public governance levers that aim to align government practice with OECD AI principles [OECD/LEGAL/0449] and ultimately increase confidence in how public institutions use AI (OECD, 2025[1]) These levers fall into three groups:
Enablers make responsible adoption possible and includes governance capacity, data and digital infrastructure, skills, financing, procurement, and partnerships.
Guardrails set expectations for transparency and accountability through rules, guidance, oversight, and evaluation mechanisms.
Engagement with stakeholders, civil society and the public ensure deployment of AI remains human-centered and responsive.
Evidence from the OECD Trust Survey enables us to assess an underlying proposition of this framework: stronger public governance, and positive perceptions of it, are associated with trustworthiness and more optimistic views of government use of AI. If this relationship holds, it suggests the existence of a virtuous cycle by which better governance supports more trustworthy AI deployment, which can in turn improve outcomes for people and strengthen their broader perceptions of government competence and trust.
Regression analysis of the 2025 Trust Survey results points to several consistent governance drivers of optimism about AI in the public sector (Figure 5.7).4 Importantly, the model includes all public governance dimensions measured in the survey, allowing the analysis to identify which dimensions are most strongly associated with higher optimism in AI public sector use across the full set of governance variables, not a pre-selected subset.
The most important variable by far is confidence that government will regulate emerging technologies appropriately to help businesses and citizens use them responsibly. Other aspects of public governance tied to stewardship around long term issues, such as balancing the interests of different generations, reducing greenhouse gas (GhG) emissions, and cooperating with a wide range of stakeholders are also strongly associated with more positive perceptions of AI. Confidence in the government’s use of personal data, trust in the national civil service and especially in the national government likewise seem to lead to more optimistic views of AI use in the public sector, however to a lesser degree. This nevertheless confirms the potential for a virtuous cycle.
Coefficients from a regression of the number of government AI items seen with confidence on perceptions of public governance and individual characteristics
Note: The figure shows selected results of an ordinary least squares regression of the number of items that individuals view with confidence in response to the question “Thinking about how government agencies could use AI in the future, how confident are you that government agencies will achieve the following outcomes.”. A confident assessment is an answer of 6-10 on the 0-10 scale. The maximum number of items is six. The regression includes perceptions of all public governance dimensions covered in the survey and trust in the national government and civil service, all coded into the categories ‘Likely/Satisfied/Confident/High or moderately high trust’ (responses 6-10 on the 0-10 scale), ‘neutral’ (response 5), ‘Unlikely/unsatisfied/not confident/Low or no trust’ (responses 0-4) and do not know. The regression coefficients of all public governance drivers where the coefficient was equal to or above 0.2 are shown. The regressions control for individual demographic characteristics, educational attainment, AI knowledge and country and is based on data for the participating OECD countries.
Source: Estimates based on the OECD Survey on Drivers of Trust in Public Institutions 2025.
5.4.1. Confidence in the government’s regulatory approach for emerging technologies is associated with positive perceptions of AI use in the public sector
There is a significant relationship between confidence in the government’s regulatory approach for emerging technologies and positive perceptions of AI use. Individuals who believe it is likely that the government will regulate new technologies appropriately to help individuals and businesses use them responsibly5 are confident in 0.75 more items out of six than otherwise similar individuals do (Figure 5.7). Survey participants could interpret effective regulations to be ones that are risk-based, proportionate, and designed to maximise benefits and promote innovation while mitigating potential harms. This relationship demonstrates a larger effect than socio-economic or demographic characteristics, and of any other public governance variable, making perception of regulatory activity the variable with the greatest effect on attitudes towards AI adoption.
In the face of this developing architecture, across the OECD, 42% of people believe it is likely that governments will appropriately regulate new technologies to help businesses and individuals use them responsibly (see Figure 5.8). In most of the participating OECD accession candidate countries with the exception of Brazil, the share is lower. This figure has on average remained stable since 2023, even though the question wording has very slightly changed6. However, confidence levels vary significantly across countries. More than half of respondents express optimism in Australia and Korea (54%), Finland (56%), Mexico (55%) and Switzerland (60%), while confidence is considerably lower in Estonia and Slovenia (33%), Ireland (30%), Latvia (26%), and the Slovak Republic (28%). Between 2023 and 2025, several countries saw notable improvements of 5 percentage points or more, including Australia, Luxembourg, Portugal, Switzerland and the United Kingdom, with Korea showing the largest increase at 12 percentage points.
Share of the population who find it likely that their government will appropriately regulate new technologies to help businesses and citizens use them responsibly, OECD, 2025
Note: The figure presents the within-country distributions of responses to the question “If new technologies (for example artificial intelligence) became available, how likely do you think it is that the federal/central/national government would regulate them appropriately to help business and citizens use them responsibly?”. The “likely” proportion is the aggregation of responses from 6-10 on the scale; “neutral” is equal to a response of 5; “unlikely” is the aggregation of responses from 0-4; and “Don't know” was a separate answer choice. “OECD” presents the unweighted average of the weighted OECD country averages.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
5.4.2. Confidence in government stewardship on long term issues is associated with positive perceptions of AI use in government
Confidence in the ability and willingness of public administrations to steward their country on long term issues is also associated with optimism towards AI use (Figure 5.7). Individuals who feel their country will reduce GhG emissions over the next ten years, select 0.39 more positive AI expectations out of six; those who are confident in the government’s ability to cooperate with stakeholders on long term issues, such as private sector organisations and trade unions, and to balance the interests of current and future generations, select 0.38 and 0.4 more items, respectively. At present, the share of the population confident in these aspects of public governance stands at 38%, 37% and 34% respectively.
A plausible reading of these results is that governance of long-term issues such as balancing the needs of current and future generations and reduction of GhG emissions and AI governance are policy issues that share structural features, such that confidence in the public administration’s capacity to manage one is logically aligned with confidence in its capacity to manage the other. In this sense, perceptions may be driven less by issue-specific performance and more by a general assessment of whether government can credibly coordinate across borders and stakeholders while acting on problems that are either long-term (as in mitigation of GhG emissions and other issues with potentially long-ranging consequences) or fast-moving (as in AI).
5.4.3. Confidence in the government’s use of personal information is associated with positive perceptions of AI use in the public sector
Trust in government data handling is also associated with higher confidence in AI deployment within the public sector. 2025 Trust Survey findings indicate that individuals who find it likely that government institutions use personal data only for legitimate purposes are more optimistic about AI implementation, selecting 0.25 more positive items when asked about potential benefits of government AI use than people who find it unlikely (see Figure 5.7). This gap exceeds demographic divides, such as 0.21 more items selected by men than by women and 0.12 more items selected by those with a post-secondary degree compared to those who did not graduate from upper secondary education. It is however smaller than the difference in items selected by age (0.35 more items selected by 18–29-year-olds than by 50+), and AI familiarity (0.74 more items selected by those who feel they could explain AI to others compared to those who do not know what it is). These findings suggest that building public trust in how governments manage personal data could be a small but effective lever for improving attitudes toward AI adoption in public services.
Data serves as a critical input for AI systems, with performance directly tied to the quality and quantity of training data. This implies that AI’s growth in government is inseparable from an increased need for safe data sharing and solid data gathering. However, this expansion introduces significant tensions: data may be incomplete, poorly structured, or inconsistently digitised, reflecting broader quality challenges. In addition to quality concerns, governance frameworks for data sharing and management are often outdated or fragmented, leading to non‑strategic and siloed approaches (OECD, 2025[1]). Lastly, governments face complex challenges in safeguarding privacy, preventing skewed data that can lead to adverse outcomes for some individuals or groups, and ensuring data security while overcoming operational hurdles (OECD, 2024[11]).
Public perception of government use of personal data remains moderately positive across OECD member countries. In 2025, 52% of people found it likely that government agencies use personal data only for legitimate purposes (see Figure 5.9). This share is 20 percentage points higher than the proportion who believe governments will protect personal information from unauthorised access or misuse when using AI specifically. (see Figure 5.2).
Confidence in this aspect of public governance varies widely by country, with particularly positive perceptions in Finland (79%) and much lower levels of confidence in several countries, notably Colombia (34%), Chile (37%), the Slovak Republic (38%), and Greece (39%). In the participating OECD accession candidate countries, the share who are confident range from 29% in Peru to 45% in Romania. For the 18 OECD countries for which this indicator is available for all three survey waves, the share has essentially remained stable. Despite overall stability, a few countries recorded notable gains since 2023. These include Australia, Canada, Costa Rica, Mexico and Switzerland, where the share of people confident in their government’s use of personal data increased by 5 percentage points or more. Japan saw the largest improvement (compared to 2021), rising from 30% to 44%. Colombia, Estonia, and Greece on the other hand witnessed a decline of 5 percentage points or more.
Share of population who find it likely or unlikely that a public agency would use their personal data for legitimate purposes only, 2025
Note: The figure presents the within-country distributions of responses to the question “If you shared your personal data with a public agency/office/department, how likely do you think it is that it would be used for legitimate purposes only?”. The “likely” proportion is the aggregation of responses from 6-10 on the scale; “neutral” is equal to a response of 5; “unlikely” is the aggregation of responses from 0-4; and “Don't know” was a separate answer choice. “OECD” presents the unweighted average of the weighted OECD country average responses.
Source: OECD Survey on Drivers of Trust in Public Institutions 2025.
Evidence from the OECD Trust Survey therefore suggests that data governance is essential not only for deploying AI systems effectively, but also for building public trust in their use. The OECD’s Governing with AI report outlines a range of tools to support this, drawing on the OECD Framework for Data Governance in the Public Sector, which can be adapted for data used in AI systems (OECD, 2025[1]).
To ensure that governments use AI systems built on high-quality data which can be scaled beyond pilots and small experiments, a clear strategic vision and associated implementation and monitoring activities are essential. Yet the 2023 OECD Digital Government Index finds that only 59% of OECD countries have a public-sector data strategy, and even fewer provide actionable implementation guidance or have the capacity to implement their vision (OECD, 2024[12]; OECD, 2026[13]). In addition to a robust strategic framework, binding instruments are essential to orchestrate and accelerate data collection and exchange between public institutions, while safeguarding individual rights and privacy. Clear, up-to-date, and enforceable legislation on data governance and personal data protection is therefore vital to deploying trustworthy AI systems at scale. Beyond legal frameworks, governments can also use non-binding policy instruments to strengthen data governance and protection (see Box 5.4).
Norway regulatory sandbox for AI and data protection
Norway has set up a regulatory sandbox on AI and data protection. It embeds “Data Protection by Default and Design” in AI innovation that uses personal data, encouraging PETs. Lessons from sandbox projects feed into practical guidance and capability-building at Datatilsynet.
France Privacy Award
Since 2016, France’s CNIL and Inria have run the CNIL–Inria Privacy Award to encourage research in data protection and privacy. Papers are selected for scientific excellence and societal impact. Recent editions have highlighted AI-relevant topics such as algorithmic transparency, PETs, and anonymisation risks.
Estonia Ministry of Economic Affairs and Communications “Bürokratt” procurement process
Estonia shows how procurement can steer privacy-preserving AI. In 2022, the Ministry of Economic Affairs and Communications launched “Bürokratt,” an AI-based virtual assistant for public services. Its procurement has prioritised PETs such as federated learning and synthetic data to reduce privacy risks.
Korea’s Personal Information Protection Commission guide for personal information processing and AI development
Korea’s Personal Information Protection Commission has issued a guide on personal information processing and AI development (2023). It clarifies legal bases for processing, sets safety expectations, and outlines measures to protect individuals’ rights in AI systems.
Source: (OECD, 2025[14]; OECD, 2025[1])
Overall, the 2025 OECD Trust Survey shows that public attitudes toward government use of artificial intelligence are marked by both cautious optimism and some reservations. While many people recognise AI’s potential to improve the efficiency and responsiveness of public services, confidence remains significantly lower when it comes to safeguarding public sector values such as oversight transparency, fairness, and data protection. These concerns are not evenly distributed: scepticism is highest among groups that already face structural vulnerabilities or have lower levels of familiarity with AI, underscoring the need for human‑centred approaches to AI governance.
The evidence also highlights a clear link between broader perceptions of public governance and optimism toward public‑sector AI adoption. Strengthening these governance foundations can therefore serve as a double dividend: improving the trustworthiness of AI systems while reinforcing overall confidence in public institutions.
While government institutions in different countries are proceeding at varying paces in leveraging AI, results from the OECD Trust Survey suggest that regardless of the level adoption, a few guiding principles could support people’s perception of trustworthiness of government uses of AI:
Embedding and investing in core public sector values and safeguards throughout the AI lifecycle could help support people’s confidence that the government’s use of AI is transparent and fair.
Investments in transparency, better public communication and the recognition of different attitudes toward (and levels of knowledge of) AI could help reduce scepticism towards the government’s use of it.
Focus on practical, user-centred applications while maintaining accessible alternatives:
Prioritise AI uses that improve service quality and responsiveness in ways that are visible to users, while maintaining non-digital and human-supported channels for those who prefer these modes of access.
Evaluate and communicate impacts on service outcomes, equity and user experience.
Maintain an adaptive governance approach for emerging technologies, grounded in public interest objectives and long-term stewardship, so that AI use can improve public governance while also remaining consistent with evolving risks, societal expectations, and regulatory practice.
Given that personal data protection is an important a driver of trust in public institutions, data governance must continue to be a foundation for the trustworthy use of AI by the public sector, and safeguards well communicated to the public.
References
[5] Adam, M. (2023), Artificial intelligence, democracy and elections, https://www.europarl.europa.eu/thinktank/en/document/EPRS_BRI(2023)751478.
[2] Government of Australia (2023), Long-Term Insights Briefing, https://www.pmc.gov.au/sites/default/files/resource/download/ltib-report-how-might-ai-affect-trust-ps-delivery.pdf.
[4] Government of France (2026), Découvrez les nouveaux engagements du service public | Services Publics+, https://www.plus.transformation.gouv.fr/ (accessed on 27 February 2026).
[7] Growns, B. et al. (2024), “The Post Office Scandal in the United Kingdom: Mental health and social experiences of wrongly convicted and wrongly accused individuals”, Legal and Criminological Psychology, Vol. 29/1, pp. 17-31, https://doi.org/10.1111/lcrp.12247.
[8] Kanzola, A., K. Papaioannou and P. Petrakis (2024), “Unlocking society’s standings in artificial intelligence”, Technological Forecasting and Social Change, Vol. 200, p. 123106, https://doi.org/10.1016/j.techfore.2023.123106.
[9] Kim, S. and Y. Lee (2023), “Investigation into the Influence of Socio-Cultural Factors on Attitudes toward Artificial Intelligence”, Education and Information Technologies, Vol. 29, https://link.springer.com/article/10.1007/s10639-023-12172-y.
[10] Kovačević, A. and E. Demić (2023), “The Impact of Gender, Seniority, Knowledge, and Interest on Attitudes to Artificial Intelligence”, IEEE Access, Vol. 12, https://doi.org/10.1109/access.2024.3454801.
[6] OECD (2026), AIM: AI Incidents and Hazards Monitor, https://oecd.ai/en/incidents?search_terms=%5B%5D&and_condition=false&from_date=2020-02-16&to_date=2026-02-16&properties_config=%7B%22principles%22:%5B%5D,%22industries%22:%5B%22Government,%20security,%20and%20defence%22%5D,%22harm_types%22:%5B%5D,%22harm_.
[13] OECD (2026), OECD Going Digital Toolkit, https://goingdigital.oecd.org/indicators (accessed on 27 February 2026).
[1] 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.
[3] OECD (2025), More Effective Social Protection for Stronger Economic Growth: Main Findings from the 2024 OECD Risks that Matter Survey, OECD Publishing, Paris, https://doi.org/10.1787/3947946a-en.
[15] OECD (2025), OECD Regulatory Policy Outlook 2025, OECD Publishing, Paris, https://doi.org/10.1787/56b60e39-en.
[14] 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.
[12] OECD (2024), “2023 OECD Digital Government Index: Results and key findings”, OECD Public Governance Policy Papers, No. 44, OECD Publishing, Paris, https://doi.org/10.1787/1a89ed5e-en.
[11] 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.
← 1. The questions on AI knowledge and AI use by governments were not asked in Brazil and Peru due to the earlier implementation of the survey in these countries.
← 2. Use cases of AI are evolving rapidly. Findings from a Global Call for Governing with AI (https://oecd.ai/call-ai-in-gov), launched by the OECD in early 2026, may show that the volume and distribution of use cases has already changed since the publication of the Governing with AI report. These cases will be published on the OECD.AI Policy Observatory.
← 3. The wording of the explanation is as follows: “Artificial Intelligence (AI) refers to computer systems that can perform tasks typically requiring human intelligence, such as understanding language, learning from data, or making decisions. You may be familiar with AI in everyday tools like smart assistants (e.g., Alexa, Siri) or personalized online recommendations. Governments also increasingly use AI for various purposes. The following question will ask about your opinions on government use of AI. Please consider both potential benefits and risks in your responses.”
← 4. The analysis is an ordinary least squares regression. The dependent variable is the count of the number of items of the potential AI outcomes variable that individuals see positively, ranging from 0 to 6. The explanatory variables are trust in the national government and national civil service, the perceptions of the different public governance dimensions and satisfaction with administrative services. Control variables are knowledge of AI, demographic characteristics, education and country fixed effects.
← 5. To learn more about the OECD's work on regulation, see, for example, the latest Regulatory Policy Outlook (OECD, 2025[15]).
← 6. The question wording changed from “If new technologies (for example artificial intelligence or digital applications) became available, how likely do you think it is that the federal/central/national government will regulate them appropriately and help businesses and citizens use them responsibly?” in 2023 to “If new technologies (for example artificial intelligence) became available, how likely do you think it is that the federal/central/national government would regulate them appropriately to help business and citizens use them responsibly?”