Governing AI in the Public Sector: Insights from the OECD 2025 Report

Artificial intelligence is no longer a speculative technology for governments - it is a working tool. From predictive analytics in social services to automated permit processing, public-sector AI systems are reshaping how citizens interact with the state. Yet as adoption accelerates, so too do questions of accountability, ethics, and oversight. The OECD’s 2025 report, Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, provides one of the most comprehensive overviews to date of how governments are deploying AI and what governance challenges persist.

Why Government Use of AI Matters

Public institutions hold unique responsibilities compared with private innovators. When AI systems influence welfare benefits, policing, or tax enforcement, errors or biases can have direct implications for fairness and public trust. Governments are therefore both regulators and role models: how they use AI sets standards for ethical technology across society. The OECD’s study - covering 200 AI use cases across 11 government functions - underscores this dual role, illustrating both promise and pitfalls.

Key Findings from the OECD Report

The report identifies three primary goals for AI in the public sector: automating or personalizing services (57% of use cases), enhancing decision-making and forecasting (45%), and enabling external use of government AI systems (only 4%) OECD, 2025. These figures reveal an emphasis on internal efficiency rather than open innovation. Governments tend to use AI for internal management, not to share models or data with the public.

Another striking observation is that AI adoption remains uneven. While national governments lead in data-driven policymaking, local administrations often lack technical capacity or budget to deploy AI safely. Only a small subset of governments have established centralized AI governance offices, and fewer still perform systematic algorithmic impact assessments.

Areas of Strength: Automation, Decision Support, and Service Delivery

In many cases, AI has improved operational efficiency. For instance, tax authorities are using machine learning to detect anomalies in filings, reducing fraud and administrative costs. Public health agencies employ predictive models to forecast disease outbreaks or resource needs. These applications show that well-targeted AI can extend public value - especially when paired with human oversight.

AI also enhances decision-support systems. Forecasting tools help agencies allocate budgets more effectively or anticipate social-service demand. In Estonia, for example, algorithmic tools support digital service delivery while maintaining citizen control through transparent interfaces. These cases demonstrate how automation, when responsibly implemented, can reinforce - rather than replace - human judgment.

Gaps and Risks: Limited Public Allowance and Weak Evaluation Frameworks

Despite progress, the OECD warns that many governments still lack robust mechanisms for evaluating AI performance and societal impact. Few maintain public registries of deployed algorithms or require post-deployment audits. The limited public allowance - where external researchers or civil society can access models or datasets - further restricts accountability. Without transparency, even well-intentioned systems risk reproducing biases or undermining citizens’ rights.

Data governance is another persistent challenge. Governments collect vast amounts of sensitive information but often lack secure, interoperable infrastructure to support responsible AI use. The report calls for stronger data-sharing agreements and ethical standards to prevent misuse.

Policy Implications: What Governments Should Do Next

The OECD recommends several steps to strengthen public-sector AI governance:

• Institutionalize oversight: Create dedicated AI governance bodies with clear mandates for ethics, risk assessment, and compliance.

• Mandate transparency: Establish registries of AI systems used in government and require algorithmic impact assessments.

• Invest in capacity building: Train civil servants in AI literacy, ethics, and data management.

• Encourage collaboration: Partner with academia and civil society to co-develop guidelines and evaluation methods.

• Promote open innovation: Share anonymized datasets and models where possible to stimulate responsible research and public trust.

These actions align closely with emerging frameworks such as the EU AI Act and the NIST AI Risk Management Framework, both of which emphasize accountability, documentation, and proportional oversight.

Conclusion: The Road Ahead for Trustworthy Public-Sector AI

The OECD’s findings make one point clear: governments are at a turning point. AI can deliver more efficient, adaptive, and personalized public services - but only if governance keeps pace. Without robust oversight, data protection, and public engagement, technological ambition may erode trust rather than build it.

A responsible approach to AI in government requires more than compliance; it demands culture change. Policymakers must view AI not just as a technical asset but as a civic responsibility. If governments can align innovation with integrity, the result will be not only smarter administration but stronger democracy.

References

• OECD (2025). Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions. Link

• European Commission (2024). EU AI Act. Link

• NIST (2023). AI Risk Management Framework. Link