Zero Trust Policy Model (ZTPM) for Agentic Cyber-Physical Systems: 25 Typed Primitives for Robot-Controlling Multi-Agent Security
arXiv·medium signal
Analyzes the threat landscape of multi-agent LLM systems controlling industrial robots through natural language via Cobot-Claw (a deployed UR3e robotic arm system). Identifies five attack classes specific to agentic cyber-physical systems and proposes ZTPM with 25 typed primitives across five enforcement layers. First concrete policy model for the AI-agents-controlling-robots threat surface.