Artificial Intelligence and Financial Accountability in the Public Sector: Assessing Fraud Detection and Error Reduction in Nigeria's Federal Institutions

Public financial reports can fail through honest error as well as through deliberate manipulation, and both weaken accountability. This study examines whether documented adoption of artificial intelligence (AI)-based fraud detection systems is associated with error reduction and financial accountability in selected Nigerian federal public institutions. Drawing on a balanced panel of five purposively selected institutions observed from 2015 to 2024 (50 institution-year observations), the study codes official documents on a five-point (0–4) rubric and estimates pooled, fixed-effects, trend-adjusted, first-difference and lagged regressions, interpreted through agency theory. AI-based fraud detection is positively and significantly associated with error reduction and accountability (pooled β = 0.826; fixed-effects β = 0.865; within R² = 0.916). The association persists, but is smaller, after controlling for a common time trend (β = 0.560) and in first differences (β = 0.542), suggesting that part of the apparent effect reflects broader, simultaneous improvements in control environments. The findings support expanded and carefully governed use of AI-assisted anomaly detection in federal financial management, provided it is embedded in a functioning chain of review, triage and investigation, while cautioning against reading the coefficients as causal effects or generalising them to the wider federal sector