Decentralized Access Control for Agentic AI in Critical Infrastructure

A recent study published on arXiv introduces a decentralized, multi-tiered access control framework designed for agentic AI systems within essential cloud infrastructures. The researchers contend that conventional role-based access control (RBAC) frameworks fall short due to the unpredictable nature of AI agents. This new architecture features four key innovations: a compound identity model linking agent actions to human authority, a hierarchical permission structure with five levels of granularity ranging from global platform access to specific parameter restrictions, a decentralized model for policy ownership allowing tool teams to manage authorization boundaries independently, and a system of progressive trust escalation complemented by safety interlocks. The study tackles critical security issues related to the deployment of autonomous AI agents in operational settings.