v3.0 Agentic artificial intelligence systems introduce autonomous computational entities capable of planning, tool usage, and multi-step reasoning. While these systems enable powerful automation across domains such as software engineering, financial systems, and infrastructure management, they also introduce novel governance risks, including delegated authority misuse, cascading system effects, and emergent multi-agent behavior that are not adequately addressed by existing AI governance approaches. This work introduces Hybrid Semantic Governance, a unified architectural paradigm for governing agentic AI systems. Within this paradigm, semantic authorization functions as a decision mechanism that evaluates agent actions based on identity, action sequences, system state, and predicted systemic impact. A complementary semantic risk modeling framework enables formal evaluation of action consequences, while delegated authority collapse is identified as a key systemic failure mode in multi-agent environments. The proposed Semantic Agent Governance Platform (SAGP) implements these concepts as a governance control plane, integrating probabilistic AI models with semi-causal reasoning mechanisms and rigid governance constraints. This layered architecture enables interpretable, enforceable, and adaptive governance of autonomous agent behavior. The paper further introduces Systemic Denial-of-Service (SDoS) as a generalized governance phenomenon arising from endogenous system overload, extending classical security models into institutional and decision-system contexts. The proposed framework provides a scalable and practical foundation for deploying governed autonomous systems across complex socio-technical environments.
Das et al. (Sun,) studied this question.