A systematic review and Zero Trust governance framework for agentic UAV robotics in public safety

Abstract

Autonomous Unmanned Aerial Vehicle (UAV) robotic systems are increasingly deployed in safety-critical public-safety environments, including emergency response, search and rescue, infrastructure inspection, traffic incident assessment, and drone-as-first-responder (DFR) operations. At the same time, the supporting AI is shifting from scripted remote control toward agentic behavior: perceiving, planning, invoking external tools (GIS, CAD), retaining mission context, coordinating with other agents, and recommending actions to human operators. This evolution multiplies the attack surface of public-safety UAV robots across identity, device, communication, data, AI-reasoning, and governance planes, while existing studies often treat UAV autonomy, cybersecurity, and governance as separate problems. This paper presents a systematic review of 2014–2026 literature spanning agentic UAV autonomy, Zero Trust Architecture (ZTA), UAV cybersecurity, LLM-agent security, and public-safety robot governance, synthesized into the Zero Trust Agentic Drone Public-Safety (ZTADP) Framework: a six-layer model—Mission, Identity & Access, Device, Communication, Agentic Reasoning, and Evidence & Governance—applying continuous verification to every actor, data stream, tool call, and AI-mediated recommendation in a UAV mission. The paper contributes a mission-pipeline threat model, design principles for least-privilege autonomy and human-confirmed action, a framework architecture, and a validation roadmap. We argue that trustworthy public-safety UAV robotics requires verifying not only who controls a robot, but also whether its own reasoning, tool use, and evidence handling can be trusted. The paper closes with a research agenda toward Zero-Trust-by-design, explainable, privacy-preserving, sustainable, and audit-ready agentic aerial robotics.