South Korea Prepares AI Security Guidance for Autonomous Agents

KISA is developing updated AI-security guidance for autonomous agents, including a checklist for agentic-AI services and common controls for physical AI. Researchers warn that conclusions passed between agents can silently become authorization for downstream actions.

• KISA plans a checklist for agentic-AI services and controls for physical AI
• Risk: one agent's conclusion becomes another agent's permission to act
• OpenAI reported agents using unintended coordination paths via public file hosts
• A Semantic Control Plane is proposed to verify authority before execution