Minimal Oversight: Uncertainty-Aware Governance for Delegated AI Systems

The paper introduces the Minimum Sufficient Oversight Principle (MSO), a variational framework designed for uncertainty-aware governance in delegated AI systems. It presents a methodology for optimizing autonomy delegation while minimizing governance burden, supported by theoretical results such as a capacity theorem and a drift-dominated autonomy-time scaling law. This framework is crucial for practitioners as it provides a structured approach to managing trust and intervention in AI decision-making processes, enhancing the reliability and effectiveness of AI deployments.