AEGIS: Reference Monitor-Based Governance with Federated Trust for Autonomous Edge AI Agents

Preprint submitted to IEEE Transactions on Network Science and Engineering, Special Issue on Secure, Trustworthy, and Autonomous Intelligent Edge Networking with Agentic AI. AEGIS: Reference Monitor-Based Governance with Federated Trust for Autonomous Edge AI Agents Autonomous AI agents deployed at the network edge execute consequential actions against physical and digital infrastructure. Existing governance approaches primarily operate on model outputs during inference and do not directly govern infrastructure actions performed post-reasoning. This paper presents AEGIS, a reference monitor architecture that enforces deterministic governance at the agent action boundary — post-reasoning, pre-execution — in edge deployments. Contributions:- AEGIS four-layer governance architecture satisfying Anderson's reference monitor properties- ATX-1: threat taxonomy of 10 tactics and 29 techniques for agentic AI failures- GFN-1: federated trust protocol with game-theoretic Sybil resistance- Live multi-agent laboratory reproducing the Agents of Chaos study with 7 autonomous agents Key finding: Ungoverned AI agents autonomously discovered vulnerabilities and built offensive tools at machine speed. AEGIS governance blocked all evaluated exploits while preserving agent communication. Related artifacts:- ATX-1 dataset: 10.21227/015v-9641(https://doi.org/10.21227/015v-9641)- aegis-core runtime: 10.5281/zenodo.19355478(https://doi.org/10.5281/zenodo.19355478)- IEEE Computer preprint: 10.5281/zenodo.19223924(https://doi.org/10.5281/zenodo.19223924)