Measurement of Corruption with AI: Beyond Hype, Toward Evidence and Enduring Statistical Challenges

This paper critically examines the promises and pitfalls of AI-based corruption measurement by situating current developments within long-standing debates in statistics, measurement theory, and data governance. We argue that many challenges attributed to AI—bias, opacity, trade-offs between accuracy and interpretability, and questionable generalizability—are not fundamentally new, but rather amplified versions of enduring statistical problems. Corruption data are often incomplete, politically skewed, non-randomly selected, and shaped by enforcement practices, media attention, and reporting incentives. When AI systems are trained on such data, they risk reproducing and legitimizing existing measurement biases under a veneer of technical objectivity. Drawing on case examples and regulatory frameworks such as the EU AI Act and OECD AI Principles, the paper highlights the governance implications of AI-based anticorruption tools. We emphasize the necessity of human oversight, algorithmic auditing, and robust data governance as preconditions for ethical and effective AI use. The contribution advances a cautious, evidence-based perspective that tempers technological enthusiasm with methodological rigor, arguing that AI can enhance corruption measurement only insofar as its statistical foundations and institutional embedding are explicitly addressed.