The Role of Artificial Intelligence in Fraud Detection: A Theoretical Examination of Financial Security, Risk Management, and Corporate Governance

Abstract

Artificial intelligence (AI) has become central to how organizations detect financial fraud, yet the theoretical literature addressing this shift remains split across three largely separate conversations: technical work on detection accuracy, governance work on organizational oversight, and an emerging and considerably more troubling literature on AI's use by fraud perpetrators themselves. This paper offers a theoretical examination that integrates these three conversations around a single argument: AI does not simply improve fraud detection as a technical capability; it restructures the relationship between financial security, enterprise risk management, and corporate governance in ways that classical fraud and agency theory did not anticipate and do not fully explain. Drawing on fraud triangle and fraud diamond theory, agency theory, and enterprise risk management frameworks, alongside recent empirical evidence on machine learning applications in financial statement fraud, anti-money laundering, and synthetic-identity crime, the paper develops an account of fraud detection as a genuinely adversarial and co-evolutionary process rather than a static technical problem to be solved once. It finds that AI simultaneously strengthens detection capability, in the form of demonstrably superior fraud- and misstatement-prediction models, and destabilizes the assumptions underlying that capability, because the same generative technologies improving detection are available to, and increasingly used by, the perpetrators detection systems are built to catch. The paper argues that this dual-use reality has direct implications for corporate governance, since boards and audit committees can no longer treat AI-enabled fraud detection as a static control to be implemented once and monitored passively, but must instead govern an ongoing arms race whose terms shift as the underlying technology develops on both sides.