KISA Publishes AI Red Team Guide

The Ministry of Science and ICT and the Korea Internet & Security Agency (KISA) have published a 69-page Korean-language AI security red-teaming guide that examines AI models and services from an attacker's perspective.

The guide organizes the AI red-teaming process into the following 3 stages:

• Preparation: Forming the red team, setting the scope, objectives, and attack paths
• Execution: Checking for AI system vulnerabilities and security risks
• Result Reporting: Summarizing findings, mitigation measures, and red-teaming deliverables

The document explains the concept and necessity of AI red teams, organizational structure, and operational procedures. The appendix includes a practical checklist, a comparison of open-source inspection tools and datasets, a job description (JD) for red team roles, and a result reporting template.

It references the international standard draft ISO/IEC AWI TS 42119-7 and cites NIST AI 100-2, the OWASP Generative AI Red Teaming Guide, and the CSA Agentic AI Red Teaming Guide as practical standards. KISA also released a response manual covering AI threat scenarios in 8 sectors, including finance, healthcare, public, education, manufacturing, and telecommunications, during the same period.