Agentic AI Security and Control Plane Evolution

This technical report discusses the evolution of security measures in the context of agentic AI systems, which are autonomous agents capable of executing tasks and interacting with various tools. It outlines the shift from traditional human-centric security frameworks to agent-native architectures that address the unique challenges posed by these systems. The report identifies six critical trust problems that arise with agentic AI and proposes solutions that include the adoption of Model Context Protocol (MCP) and A2A frameworks. It emphasizes the need for a data-centric trust architecture that supports real-time authorization and auditing for AI agents, moving away from session-based controls. Additionally, the report details the implementation of Named Data Networking (NDN) as a protocol-gapped trust fabric, enhancing security through cryptographic measures and strict access controls. The findings highlight the importance of adapting security practices to accommodate the complexities of multi-agent workflows and the necessity of evolving identity and access management strategies for AI systems.