South Korea Drafts Security Rules for Autonomous AI Agents

KISA, under the Ministry of Science and ICT, is developing version 2.0 of its AI Security Guide covering autonomous AI agents in software, networks and physical systems. The document assigns responsibility: developers limit agent tool access and keep secure decision logs, providers ensure emergency shutdown, and companies set permissions and require human approval for risky actions.

• KISA prepares AI Security Guide 2.0 for autonomous AI agents
• Developers to limit agent tool access and keep decision logs
• Providers to add emergency shutdown and incident tracking
• Human approval required for risky actions