CERT-In issues AI Security Auditing Guidelines with a risk-tiered framework and 6-hour incident reporting
The Indian Computer Emergency Response Team (CERT-In) released AI Security Auditing Guidelines on 4 November. They set out how organisations should audit AI systems, including agentic AI, which must be tested for unsafe tool use, excessive autonomy and privilege misuse.
Why in news
Background: The guidelines apply to organisations that develop, buy, deploy or operate AI systems in India, and to AI systems that process Indian data or serve Indian users. They come amid incidents where AI agents accessed live systems of Indian organisations.
Prelims pointers
- Nine audit domains cover the AI lifecycle, including governance, data and privacy, model risk, bias and fairness, explainability, security, monitoring, human oversight and supply-chain assurance.
- AI systems are classified into four risk tiers: Low, Moderate, High and Critical. The tier decides how deep, how frequent and how independent the audit must be.
- Moderate-risk systems need annual independent audits by CERT-In-empanelled auditors. Critical systems need the highest scrutiny and continuous monitoring.
- Three human oversight models are defined: human-in-the-loop, human-on-the-loop and human-over-the-loop.
- AI security incidents must be reported to CERT-In within six hours, and logs must be kept for 180 days.
- Agentic AI refers to AI systems that can plan and take actions on their own, such as using software tools, with limited human input.
- CERT-In is the national agency for cyber incident response under Section 70B of the IT Act, 2000.
Mains angle
As AI agents begin to act inside real systems, the risks move from wrong answers to wrong actions, such as data leaks or unauthorised changes. A risk-based audit framework is a practical step for AI governance in India. Its success depends on auditor capacity, compliance costs for smaller firms and clarity on how it fits with the IT Act and the DPDP Act.