CERT-In Agentic AI Audit Guidelines
3.6 Science And Tech, The Hindu, 2026-11-06
CERT-In has issued the AI Security Auditing Guidelines: Framework for Secure, Ethical, and Auditable Artificial Intelligence Systems for agentic AI that plans tasks, uses external tools and delegates work. Auditors must test for exceeding permissions, bypassing approvals, credential misuse, memory manipulation and cross-user data leakage. Multi-agent systems need unique agent identities, authenticated and logged communications, and controls on delegation loops. High-risk systems need emergency-stop mechanisms to suspend agents and revoke credentials. Organisations must keep an inventory of models, training data, software and tools.