AI governance drift: Why audit and governance controls fail after AI deployment

Abstract

Artificial intelligence (AI) systems are increasingly being deployed into dynamic environments in enterprise settings, where some governance controls can become lax once they have been deployed. The study proposes a new phenomenon called “AI Governance Drift,” which refers to the phenomenon of a shifted alignment between the governance mechanisms and the behavior of the deployed AI systems over time. This research is based on the concept of Dynamic Capabilities Theory. It explores how continuous audit capability, governance monitoring, assurance over cybersecurity, and governance visibility impact operational resilience in AI-enabled enterprises. Based on the survey results from 700 professionals in Vietnam and analysis using Partial Least Squares Structural Equation Modeling (PLS-SEM), the results indicate that there is strong evidence that the continuous audit approach significantly lowers governance drift, and the approach of providing visibility of governance has a positive effect on the operational resilience of organizations. The study also finds that governance visibility is an important intermediary between governance monitoring and decreased governance deterioration. The research extends governance literature to the post-deployment AI environment. It highlights the significance of adaptive and continuous assurance mechanisms in autonomous digital enterprises for AI governance and IT audit