South Korea to develop new security guidelines for autonomous AI agents

South Korea is drafting security guidelines for autonomous AI agents operating across critical industries. The Korea Internet & Security Agency, under the Ministry of Science and ICT, is developing the updated "AI Security Guide" version 2.0. The proposed rules will cover access controls, audit logs, shutdown tools, and human oversight for the deployment of agentic AI systems.

These autonomous agents, which can receive credentials and use external tools, may require tighter access controls, stronger audit trails, and human approval for consequential actions. The guide aims to address security concerns arising from the deployment of agentic AI services, providing a checklist to manage associated risks.

The agency plans to divide security responsibilities across the AI deployment pipeline, focusing on granular controls to minimize the risk of unauthorized or harmful actions by these AI agents.