Assessing Transparency and Accountability Mechanisms in AI-Driven Audit Tools

Abstract

The current paper looks at transparency and accountability features integrated into the artificial intelligence-driven audit tools with critical concerns regarding the algorithmic decision-making in the financial and regulatory oversight settings. A mixed-methods methodology that involves technical examination of fifteen popular audit systems, semi-structured interviews with sixty professionals in various stakeholder groups, and three detailed case studies, will enable us to methodically evaluate how modern AI audit systems apply explainability functionality, provide decision-making log trails, and create accountability systems. We find that transparency practices strongly vary across the various categories of tools, as commercial platforms show moderate adoption of explainable AI capabilities but accountability mechanisms are largely procedural and not technical. We outline problematic aspects of existing implementations, such as limited stakeholder access to algorithmic decision-making, a lack of algorithmic decision process provenance and model validation processes documentation, and inadequate interaction with existing regulatory frameworks for the audit practice. The research suggests a multi-dimensional framework of evaluation of transparency and accountability of AI audit tools, which is a consolidation of technical, procedural, organization, and regulatory factors. We provide evidence-based recommendations to developers, audit practitioners, and regulators to enhance trustworthiness and performance of AI-driven audit systems while addressing inherent trade-offs between transparency, competitiveness, and system complexity.