Agentic AI Risk Governance: A Policy-Driven Regulatory Framework

Agentic AI systems, designed for autonomous and adaptive decision making, raise complex challenges in regulatory compliance and ethical risk management. Operating in high-stakes domains like healthcare, finance, and autonomous systems, these models increase the likelihood of safety breaches, ethical violations, and legal misalignment. This paper introduces a policy-driven Risk Regulatory Framework (RRF), implemented through a Decision Support System that combines machine learning, explainable AI, and rule-based compliance logic. Risk classification is modeled using a Random Forest LSTM ensemble, while compliance violations are detected via a logic-weighted neural function. SHAP values provide local and global interpretability, ensuring alignment with transparency principles. Evaluated on synthetic datasets representing real-world conditions, the system achieved the highest accuracy in finance (59.7%), followed by autonomous systems <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(43.5 {\%})$</tex> and healthcare <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(37.1 {\%})$</tex>. Violation detection precision and stakeholder trust followed a similar trend. These results demonstrate the framework's potential for reproducible, explainable, and policy-aligned governance of Agentic AI.