AI in Government Statistics 2026

The governance gap: 74% adoption, 18% effectiveness

The defining tension in government AI is not whether public servants use it. They do, at rates that rival or exceed the private sector. The real question is whether anyone is steering. A landmark survey of 3,335 public servants across 10 countries found that 74% already use AI in their work, yet only 18% believe their government deploys it effectively. That 56-percentage-point gap between adoption and perceived effectiveness is the single most important number on this page.

The gap between adoption and governance runs deeper than the headline numbers suggest. In organizations where clear policies and leadership support are absent, 70% of active AI users deploy it without their employer's knowledge. And 64% of enthusiastic AI users rely on personal logins for work tasks, bypassing enterprise tools entirely. This is not cautious experimentation. It is an unmanaged transformation happening beneath the surface of government operations.

Cross-referencing government AI adoption and effectiveness data from ITIF with private-sector depth-of-transformation figures from Deloitte reveals a structural difference. In government, the ratio of reported adoption to perceived effectiveness is 4.1:1. In the private sector, the ratio of adoption to deep transformation is 2.6:1. Government AI deployments run roughly 1.6 times more shallow than their private-sector equivalents (TheAIDaily based on ITIF Public Sector AI Adoption Index 2026 and Deloitte State of AI in the Enterprise 2026).

• Where leadership is present, the picture shifts dramatically: 65% of public servants use AI frequently when management actively supports it, compared to just 37% where support is low, a 28-percentage-point gap driven by organizational culture rather than technology (Gallup, Q4 2025).
• In countries with clear AI guidelines, 91% of public servants feel confident using AI and 82% are optimistic about its impact, versus sharply lower figures where policies are absent (ITIF, 2026).
• Only 37% of public-sector employees say their organization has a clear AI strategy, compared to 53% in the private sector, a 16-point strategy gap that helps explain the shadow AI phenomenon (Gallup, Q4 2025).
• The top request from public servants is not more budget (ranked 12th) but clearer guidelines (38%), confirming the gap is governance, not money (ITIF, 2026).

Sources: ITIF/Public First, Public Sector AI Adoption Index 2026 (n=3,335, 10 countries); Gallup Workplace Survey Q4 2025; Deloitte, State of AI in the Enterprise 2026 (n=3,235, 24 countries); TheAIDaily compilations

How widely do governments use AI?

Government AI adoption is no longer a pilot-stage curiosity. Across the OECD, 97% of member countries now deploy AI in at least one government domain. In the United States, nearly 90% of federal agencies are either using or actively planning AI deployments. The question has shifted from "whether" to "how deep."

But the depth of that adoption varies sharply. According to Gallup's Q4 2025 survey, 43% of public-sector employees use AI at least a few times per year, and 21% use it daily or multiple times per week. These figures put government essentially on par with the private sector, where 41% report any AI use and 25% use it frequently. The gap between sectors is closing faster than most policymakers expected.

• Global AI market projections put AI in government and public services at $26.4 billion in 2025, growing to $160 billion by 2036 at a compound annual growth rate of 17.8% (Future Market Insights, 2026).
• Organization-level adoption worldwide reached 88% in 2026, with generative AI used in at least one business function by 70% of organizations across all sectors (Stanford HAI AI Index, 2026).
• Across OECD member countries, 20.2% of firms used AI in 2025, more than doubling from 8.7% in 2023, with large firms (52%) adopting three times faster than small firms at 17.4% (OECD, 2026).
• Only 1% of government respondents say that 60% or more of their staff have access to generative AI tools, the lowest of any sector surveyed by Deloitte (Deloitte State of AI 2026).
• Government fraud detection is one of the highest-impact applications: the US Treasury recovered over $4 billion in fraudulent payments in FY2024 using machine learning for real-time detection (US Treasury, 2024).

The SAS government fraud survey adds another dimension: approximately half of government organizations now use AI in some form, roughly a quarter use generative AI specifically, and 97% of respondents expect to deploy GenAI within two years. Network analysis for fraud detection is projected to grow from 32% to 87% adoption in the near term (SAS, May 2025).

Sources: Gallup Workplace Survey Q4 2025; OECD Digital Government Outlook 2026; Stanford HAI AI Index 2026; Future Market Insights 2026; Google Public Sector/Government Executive survey 2026; Deloitte State of AI 2026; SAS Government Fraud AI Study 2025; US Treasury FY2024

AI use cases across government functions

The OECD cataloged 200 real-world government AI deployments across its member countries in 2025, spanning 11 core government functions. The distribution reveals where governments are actually deploying AI today, and where they are not.

Half of all cataloged use cases fall into just three functions: public service design and delivery, justice administration, and civic participation. Meanwhile, policy evaluation, arguably the function where data-driven analysis could add the most value to democratic governance, has the fewest deployments at just 5 out of 200.

The objectives behind these deployments cluster around three goals: 57% of cases support automating, streamlining, or tailoring services; 45% enhance decision-making, sense-making, or forecasting; and 30% aim to improve accountability and anomaly detection. These percentages overlap because many use cases serve multiple purposes.

• Rules-based systems and selective machine learning dominate government AI today; generative AI and large language models remain rare in production deployments, though pilot projects are expanding rapidly (OECD, 2025).
• The US federal government reported 3,611 AI use cases across 56 agencies in 2025, more than doubling from 1,757 in 2024. The growth from roughly 700 in 2023 to 3,611 in 2025 represents a fivefold increase in two years (OMB Federal AI Use Case Inventory, 2025).
• Singapore's GovTech "Pair" assistant has reached 60,000+ registered users and 20,000+ weekly active users, with over 10 million messages processed. Civil servants report approximately 46% time savings on administrative tasks (GovTech Singapore, 2025).
• Finland's Kela social security institute deployed an AI platform for document classification that saves an estimated 38 full-time-equivalent years of caseworker time annually (OECD, 2026).
• The UK tested an AI consultation-analysis tool ("Humphrey") on 26 live government consultations, saving thousands of administrative hours and over GBP 500,000, in a process that previously cost roughly GBP 100,000 per consultation in external consultant fees (GOV.UK, 2025).

Generative AI use in government is growing but remains early-stage. Among OECD countries, 56% report using GenAI to support public servants, 42% for automated report generation, 42% for citizen-engagement content, and 36% for drafting policy documents (OECD Digital Government Outlook 2026).

Sources: OECD, Governing with Artificial Intelligence 2025 (200 use cases); OMB Federal AI Use Case Inventory 2025; GovTech Singapore 2025; OECD Building an AI-Ready Public Workforce 2026; GOV.UK AI Opportunities Action Plan One Year On 2025; OECD Digital Government Outlook 2026

Government AI readiness by country

Multiple global indices track how prepared governments are to adopt and govern AI. The picture they paint is consistent in some dimensions and divergent in others: wealthy democracies cluster at the top of most rankings, but regional powers with strong state-directed AI strategies sometimes outperform them on specific pillars.

The ITIF Public Sector AI Adoption Index 2026 groups countries into three tiers based on actual civil-servant adoption patterns, not just policy frameworks. The results challenge several assumptions about which governments lead in practice.

• Advanced Adopters (Singapore, Saudi Arabia, India) combine strong leadership with widespread daily AI use among civil servants. Singapore scores highest on empowerment (85% of public servants feel confident using AI) while Saudi Arabia leads on embedding (65% report access to enterprise AI tools) (ITIF, 2026).
• Uneven Adopters (US, UK, South Africa, Brazil) show strong pockets of adoption but face persistent gaps in infrastructure or policy guidance. The US scores highest on enablement (89% report organizational AI tool availability) but trails on governance frameworks (ITIF, 2026).
• Cautious Adopters (Germany, France, Japan) remain risk-averse, with AI largely confined to specialist applications. France ranks last overall: 74% of French public servants say AI cannot perform any part of their job, roughly 45% never use AI at work, and only 27% report organizational investment in AI tools (ITIF/Euronews, 2026).
• Estonia maintains 120+ AI use cases across 60+ organizations with 40+ reusable AI components under its Kratt programme. Its Buerokratt open-source state assistant was recognized among UNESCO's top 100 AI projects (Estonian RIA, 2025).
• The UAE claims 97% AI-tool utilization across government entities in 2025, with a national strategy targeting AI's share of GDP rising from approximately 9% to 45% by 2031. These are self-reported government figures and should be treated as aspirational benchmarks rather than independently verified data (UAE AI Office, 2025).

The UN E-Government Survey 2024 provides the broadest assessment, covering all 193 member states. The global average E-Government Development Index (EGDI) rose from 0.6102 in 2022 to 0.6382 in 2024, and the share of the world's population in countries lagging in digital government development dropped from 45% to 22.4%. A total of 76 countries now fall in the "very high" EGDI tier.

Sources: Oxford Insights Government AI Readiness Index 2024/2025; Tortoise Global AI Index 2024; UN E-Government Survey 2024; ITIF/Public First Public Sector AI Adoption Index 2026; Estonian RIA 2025; UAE AI Office 2025

Algorithm transparency and public AI registries

Knowing what AI systems governments use is a prerequisite for accountability. Yet the global state of algorithm transparency remains patchy. Only 3 of 36 OECD countries maintain mandatory AI use-case repositories, and even the most comprehensive registers reveal more problems than they solve.

The Netherlands operates the most developed algorithm register in the world, with 1,485 algorithm descriptions published by 515 government organizations as of mid-2026. The register became mandatory in 2025 after launching as a voluntary initiative in 2023 with roughly 1,000 entries. Approximately 100 organizations have joined the sAIdkick programme, a government-wide marketplace for generative AI tools (Digitale Overheid, 2026).

But transparency and compliance are not the same thing. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) found that more than 50 of approximately 100 selection instruments used by the Dutch Tax Authority are potentially discriminatory or unlawful. For more than half, no validated statistical analysis exists. The Tax Authority is now under enhanced supervisory oversight for five years (AP/Follow the Money, 2025).

• The Netherlands' childcare benefits scandal remains the most cited example of algorithmic harm in government: approximately 26,000 parents were wrongly flagged as fraudsters by a self-learning risk algorithm that used dual nationality as a criterion, and over 180,000 citizens were placed on an unverified blacklist (Dutch parliamentary inquiry, 2019-2021).
• Denmark deploys 60+ AI and machine learning models for welfare fraud detection through agencies Udbetaling Danmark and ATP, processing data on millions of residents. An Amnesty International investigation found that the models risk discriminating against marginalized groups, though Amnesty was only able to examine 4 of the 60+ models (Amnesty International, 2024).
• France's CNAF welfare algorithm scores 32 million people monthly and calculates over 13 million suspicion scores, with criteria such as low income, unemployment, and disability increasing a person's risk score. The algorithm's source code has been challenged in court by 25 organizations (La Quadrature du Net, 2023; Amnesty 2024).
• Across OECD countries, only 28% (10 of 36) conduct any form of financial or non-financial impact measurement of government AI use cases, whether prospective or retrospective. The remaining 72% deploy AI without systematic evaluation of its effects (OECD, 2025).

Sources: OECD Digital Government Outlook 2026; Dutch Algoritmeregister (algoritmes.overheid.nl); Autoriteit Persoonsgegevens 2025; Dutch Parliamentary Inquiry (Toeslagenaffaire); Amnesty International, Coded Injustice 2024; La Quadrature du Net 2023; Digitale Overheid (sAIdkick)

EU AI Act and public-sector compliance

The EU AI Act, which entered into force on August 1, 2024, creates the world's most comprehensive regulatory framework for government AI. Many of the Act's highest-impact provisions directly target public-sector applications: law enforcement, migration, access to essential services, and judicial decision-making are all classified as high-risk.

The compliance timeline has shifted. The original deadline for high-risk AI systems under Annex III was August 2, 2026, but the EU's Digital Omnibus package delayed this by 16 months to December 2, 2027 (provisional agreement, May 2026). Public authorities using existing high-risk AI systems receive an additional grace period until August 2, 2030, to accommodate procurement cycles and resource constraints. Prohibited AI practices, such as social scoring and certain forms of biometric surveillance, have already been enforceable since February 2, 2025.

• High-risk categories affecting government include AI systems used for law enforcement, migration and border control, administration of justice, and access to essential public services. A municipal welfare screening tool faces the same requirements as a national immigration authority (EU AI Act, 2024).
• Of 36 OECD countries, 69% (25 of 36) now have formal or binding AI requirements in place, 83% (30 of 36) use soft governance approaches such as guidelines and recommendations, and 53% (19 of 36) combine both hard and soft mechanisms (OECD Digital Government Outlook, 2026).
• Ethical guidelines for generative AI have been adopted by 75% (27 of 36) of OECD countries, and 58% (21 of 36) run specific training programs on responsible GenAI use for public servants (OECD, 2026).
• National AI legislation is accelerating globally: the number of AI-related laws in US states rose from 1 in 2016 to 131 in 2024, while 59 AI-focused federal regulations were issued in 2024 alone, more than doubling the previous year, spread across 42 agencies (Stanford HAI, 2025).
• At least 94 countries have developed or are rolling out a national AI strategy, reflecting the global consensus that AI governance cannot wait for the technology to mature (Stanford HAI, 2025).

Sources: EU AI Act (Regulation 2024/1689); European Commission Digital Omnibus 2026; OECD Digital Government Outlook 2026; Stanford HAI AI Index 2025; artificialintelligenceact.eu

Key takeaways

• Government AI adoption is high but shallow. At 74%, public-servant adoption rivals the private sector, but the 4.1:1 ratio of adoption to perceived effectiveness (vs. 2.6:1 in the private sector) indicates that most government AI use remains unstructured and informal.
• Shadow AI is the dominant mode of government AI use. Where clear policies are absent, 70% of users deploy AI without employer knowledge, and only 37% of organizations have a stated AI strategy.
• Federal AI spending is exploding, concentrated in defense. US federal AI contract spending grew 966% in two years to $7.2 billion, with 98.9% flowing to the Department of Defense. Civilian agencies lag but are scaling.
• The talent gap may be the binding constraint. Government offers the lowest AI wage premium of any sector (16% vs. 62% average), roughly half of UK government digital roles go unfilled, and the private sector posts 174 AI jobs for every 1 in government.
• Transparency is necessary but insufficient. Even the Netherlands, which leads the world with 1,485 registered algorithms, finds widespread non-compliance upon inspection. Only 3 of 36 OECD countries require registries at all.
• Country tiers defy expectations. India and Saudi Arabia are classified as "Advanced Adopters" while Germany, France, and Japan remain "Cautious," suggesting that state-directed AI strategies can outpace decentralized democratic approaches in adoption speed.
• Regulation is catching up. The EU AI Act's December 2027 high-risk deadline (delayed from August 2026) gives governments 16 more months, but Article 4's AI literacy requirement is already in effect and applies to every public-sector AI deployer.

Sources: ITIF 2026; Gallup Q4 2025; Brookings 2026; PwC 2026; OECD 2025/2026; EU AI Act; TheAIDaily compilations