CERT-In Asks Organisations to Test AI Agents for Misuse of Tools and Permissions
The Indian Computer Emergency Response Team (CERT-In) has issued Artificial Intelligence Security Auditing Guidelines, which give organisations a framework for checking how safe their AI systems are.
A special focus is on AI agents: AI software that can plan tasks, use other software tools, reach data and take actions with little human help.
Audits must check whether an agent uses tools it was not allowed to use, or misuses its permissions (privilege abuse). They must also check how agents hand over tasks to other agents, and whether a working emergency-stop ("kill switch") exists.
In multi-agent systems (where several AI agents work together), each agent should have its own identity and clearly defined permissions, and every message between agents must be authenticated and logged. Records of the actions agents take must be kept.
AI systems are to be sorted by risk into low, moderate, high and critical, based on how independently they act, how sensitive their data is, how many people they affect and whether harm can be undone. Higher risk means stricter checks, up to continuous monitoring and yearly red-team testing for critical systems.
The guidelines also expect security logs to be kept for at least 180 days with matching timestamps, and reportable AI-related cyber incidents to be reported to CERT-In within six hours, in line with its existing directions. This follows an October 2026 CERT-In advisory warning that some AI agents under testing had gone beyond their tasks and accessed live systems of Indian organisations.
CERT-In and India's Cybersecurity Framework
CERT-In, the Indian Computer Emergency Response Team, is India's national agency for handling cyber security incidents such as hacking, data breaches and malware attacks. It works under the Ministry of Electronics and Information Technology (MeitY). You can think of it as the country's "fire brigade" for cyber attacks: it warns people about new dangers, helps put out the fire when an attack happens, and issues rules so that the same fire does not start again.
The AI auditing guidelines extend CERT-In's role from ordinary software to AI agents. They reuse its existing tools, the 6-hour reporting rule and 180-day logs, so that when an AI agent misbehaves, investigators can trace what it accessed, what instruction it received and what it did.
Agentic AI: From Assistants to Autonomous Systems
Agentic AI means AI systems that do not just answer questions but act on their own to finish a task. You give such a system a goal, and it plans the steps, uses tools like web browsers, code and software, checks its own results and keeps going until the job is done. Each such system is called an AI agent. Because agents can act in the real world, a mistake or a misuse can cause real harm, not just a wrong answer.
CERT-In's guidelines target exactly these agent risks: testing for unauthorised tool use and privilege abuse, checking task hand-offs and emergency stops, and demanding authenticated, logged communication between agents so every action can be traced.
- Issuing body: CERT-In, under the Ministry of Electronics and Information Technology (MeitY)
- Document: Artificial Intelligence Security Auditing Guidelines (November 2026)
- Risk tiers: low, moderate, high, critical (critical: continuous monitoring and annual red-team testing; moderate: independent audit at least once a year)
- Multi-agent systems: unique identity and defined permissions for each agent; authenticated and logged inter-agent communication
- Security logs: at least 180 days; incident reporting to CERT-In within 6 hours (April 2022 directions)
- Legal basis of CERT-In: Section 70B, IT Act, 2000 (inserted by the IT (Amendment) Act, 2008); operational since 2004
- Earlier advisory: October 2026, on AI agents under evaluation accessing systems of Indian organisations beyond their assigned scope