Supreme Audit Institutions must evolve to govern AI in public sector

The accelerating deployment of artificial intelligence (AI) across public sector institutions is ushering in a transformative phase for government oversight and accountability systems. As public entities adopt AI to improve policy delivery, internal management, and citizen services, the role of Supreme Audit Institutions (SAIs), charged with independently monitoring the use of public funds, has become more complex and urgent. These developments compel SAIs to evolve technologically, institutionally, and methodologically.

A comprehensive new study titled "Artificial Intelligence and Public Sector Auditing: Challenges and Opportunities for Supreme Audit Institutions", published in World, explores this shifting landscape. Authored by Dolores Genaro-Moya, Antonio Manuel López-Hernández, and Mariia Godz, the paper adopts a conceptual and exploratory approach, offering an in-depth diagnosis of how SAIs are navigating AI’s dual nature: a tool of efficiency and a source of governance risk.

How is AI reshaping public sector performance and oversight?

AI is poised to revolutionize public administration on a scale comparable to past industrial revolutions. By enabling more responsive policymaking, AI allows governments to anticipate trends, streamline services, and improve operational efficiency. The study outlines five key AI-enabled improvements in the public sector: enhanced decision-making, personalized public services, streamlined internal management, anti-corruption efforts, and greater public safety. These advances directly intersect with the oversight roles traditionally held by SAIs.

However, these benefits come with significant risks that threaten the integrity of governance. Issues around privacy, data security, transparency, and algorithmic discrimination are particularly concerning. AI systems, especially those based on machine learning, can be opaque, making it difficult for both citizens and auditors to understand how decisions are made. These risks are amplified by the lack of technical expertise in public bodies and a shortage of AI-savvy talent willing to work in the public sector.

The EU’s AI Act, the UK’s Data Science Ethical Framework, and initiatives by the UN and OECD have attempted to address these governance gaps. But as the study emphasizes, institutional competence must extend beyond legal compliance. SAIs must verify that AI systems align with ethical norms and democratic accountability standards while ensuring that public sector employees are reskilled for AI-centric environments.

How are SAIs using AI in their own operations?

The study identifies several ways in which SAIs can deploy AI to enhance audit effectiveness and efficiency. AI tools can be used during audit planning to identify high-risk areas and relevant datasets. During implementation, AI allows auditors to analyze vast amounts of data without relying solely on sampling. Machine learning models, such as anomaly detection algorithms, clustering methods, neural networks, decision trees, and robotic process automation (RPA), are now increasingly integral to these processes.

For example, India’s SAI has established a Data Management and Analysis Centre to guide data-related audit activities, while the U.S. Government Accountability Office (GAO) launched an innovation lab focused on data science and AI applications. Brazil’s Federal Court of Accounts uses a robot named Alice to analyze public tenders, and the Philippines’ SAI has deployed the MIKA-EL platform for identifying anomalous transactions at scale. In Spain, AI is used to automate invoice recognition in electoral audits. These examples reveal that while progress varies widely, pioneering SAIs are leveraging AI to improve audit precision and reduce manual workload.

Nonetheless, the study highlights a crucial bottleneck: data quality. Effective AI systems require structured, reliable, and ethically sourced data. This necessitates robust data governance strategies, secure infrastructure, and institutional arrangements that support data sharing. Moreover, the establishment of internal audit platforms that integrate data collection, processing, and reporting is deemed essential for maximizing AI benefits within SAIs.

Are SAIs ready to audit AI systems in the public sector?

While AI can be a powerful audit tool, SAIs are also expected to audit AI systems used by the public sector, a far more demanding task. The study identifies this as a critical gap. Although SAIs have begun exploring AI audits, most remain in nascent stages, lacking specialized teams, audit protocols, and relevant technical expertise. The Netherlands’ SAI has audited algorithm usage in government, focusing on bias and ethical oversight, while the UK’s National Audit Office has assessed AI governance and infrastructure readiness. The European Court of Auditors has also evaluated the EU’s AI ecosystem development.

To support emerging audit capabilities, a coalition of European SAIs produced the 2020 white paper “Auditing Machine Learning Algorithms”, which provides practical guidelines for auditing AI models. It recommends focusing on transparency, fairness, explainability, and maintainability, key audit targets often overlooked in favor of performance metrics.

The path forward is complex. Barriers include auditors’ limited technical training, scarce case studies, and insufficient legal frameworks. According to a European Court of Auditors survey, common challenges faced by SAIs include low technical skills, the multidisciplinary nature of AI, legal and ethical constraints, and financial limitations.

The study recommends several strategic imperatives: building AI-focused teams, enhancing inter-institutional cooperation, developing internal AI auditing protocols, and designing a comprehensive data strategy. SAIs must also invest in continuous training aligned with models like the European Commission’s competence framework, which classifies necessary skills into attitudinal, operational, and literacy categories.