CERT-In Issues AI Security Auditing Guidelines
Why in News?
- The Indian Computer Emergency Response Team issued AI Security Auditing Guidelines on 4 November 2026, and explainers on 5 to 7 November described them, secondary reports said.
- The framework sorts AI systems into four risk tiers, Low, Moderate, High and Critical, using autonomy, impact severity, data sensitivity, operational criticality and recoverability.
- It covers nine audit areas, including governance, data privacy, model risk, fairness, transparency, cybersecurity, human oversight and third-party suppliers, and adds extra requirements for agentic AI.
Key Terminologies
- CERT-In
- The Indian Computer Emergency Response Team, the national agency under the Ministry of Electronics and Information Technology for responding to cyber incidents. It works under Section 70B of the Information Technology Act, 2000.
- Agentic AI
- AI systems that can plan and act with some independence, for example by using tools, keeping memory and taking steps without a human approving each one.
- Risk Tier
- A level that sets how strict the audit must be. Higher tiers get stricter checks because failure can cause more harm.
- Human Oversight
- Keeping a person able to review, stop or override an AI system's decisions, especially where harm is possible.
- Ministry of Electronics and Information Technology
- The ministry that oversees CERT-In and India's digital policy, including the IndiaAI Mission and the Information Technology Act, 2000.
- Empanelled Auditors
- Audit firms approved by CERT-In to carry out security audits. Independent audits under the guidelines are to be done by such firms.
- Kill Switch
- A built-in way to shut down an AI system quickly in an emergency. Reports said the guidelines expect emergency shut-off safeguards for agentic AI.
Key Issues
- Four Tiers: AI systems are classed as Low, Moderate, High or Critical, based on decision-making autonomy, impact severity, data sensitivity, operational criticality and recoverability.
- Audit Areas: Audits cover nine areas, among them governance, data quality and privacy, model risk, fairness, transparency, cybersecurity, human oversight and third-party suppliers.
- Agentic AI: Extra requirements cover tool access, persistent memory and emergency shut-offs for systems that act on their own.
- Audit Frequency: Reports differ. One says low-risk systems need an annual review, moderate ones a yearly independent audit by a CERT-In empanelled auditor, and high-risk ones audits as often as every six months.
- Incident Reporting: One outlet reports a six-hour timeline to report AI security incidents to CERT-In. CERT-In's cyber incident directions of 2022 already require reporting within six hours, so this fits, but the official text should be checked.
Key Implications
Positive/Pros/Merits
- Clear Method: A tiered approach makes audits proportionate, with the strictest checks for the riskiest systems.
- Trust in AI: Audits and human oversight help businesses and citizens trust AI used in services.
- Agentic AI Safety: Specific rules on tool access, memory and shut-offs address new risks.
- Market Standard: Empanelled auditors create a supply of qualified audit firms, which can build a domestic AI assurance industry and give buyers a common reference when they compare vendors.
Negative/Cons/Demerits
- Compliance Cost: Small firms and start-ups may find independent audits costly.
- Legal Status: Guidelines are not a law, so their binding force and penalties depend on other rules.
- Unclear Details: Outlets differ on audit frequency, so firms must wait for or check the official document.
- Fast-Moving Field: AI systems change quickly, and audit rules may lag behind new capabilities, so the guidelines will need regular review as agentic systems become more common in banking, health and public services.
Key Initiatives
- CERT-In: It issued AI Security Auditing Guidelines on 4 November 2026 with four risk tiers and nine audit areas, according to secondary reports.
- Empanelled Auditors: Independent audits are to be done by auditors empanelled by CERT-In.
- Agentic AI Safeguards: The guidelines add requirements on tool access, memory and emergency shut-offs.
- Ministry of Electronics and Information Technology: CERT-In works under it.
This article relies on news and explainer coverage and not on the official CERT-In document, which was not reached. Details on audit frequency and the six-hour reporting rule vary by source and should be checked against the official text.
Director's Perspective
Way Forward
- Know that CERT-In is the national cyber incident response agency under MeitY.
- Learn the idea of risk-based tiers and proportionate audits.
- Understand agentic AI and why it needs human oversight and shut-offs.
- Link the guidelines to the IT Act, data protection and the IndiaAI Mission.
- In answers, balance innovation with safety and accountability.
- Compare the CERT-In approach with the risk-based approach of the European Union's AI Act.
The CERT-In guidelines bring a risk-based audit approach to AI in India, with four tiers, nine audit areas and extra safeguards for agentic systems. They support trust and safety but impose costs, and outlets differ on key details such as audit frequency. In a Mains answer, explain the tiered framework and human oversight, link it to the IT Act and data protection law, and conclude that clear, proportionate rules with support for small firms will let India encourage AI while limiting harm.
GS Relevance
Frequently Asked Questions
What are the CERT-In AI Security Auditing Guidelines?
They are guidelines issued by the Indian Computer Emergency Response Team on 4 November 2026 for auditing the security of AI systems. They sort systems into four risk tiers and cover nine audit areas, with extra requirements for agentic AI, secondary reports said.
What are the four risk tiers for AI systems?
The tiers are Low, Moderate, High and Critical. They are set by decision-making autonomy, impact severity, data sensitivity, operational criticality and recoverability, so the riskiest systems face the strictest audits, according to coverage of the guidelines.
What is agentic AI?
Agentic AI refers to AI systems that can plan and act with some independence, using tools and memory. The guidelines add requirements on tool access, persistent memory and emergency shut-offs for such systems, secondary reports said.
PYQ Practice — Statement Analysis
1 CERT-In works under the Ministry of Electronics and Information Technology.
It issued AI Security Auditing Guidelines in November 2026.
2 The guidelines classify AI systems into two risk tiers.
Reports describe four tiers: Low, Moderate, High and Critical.
3 Agentic AI systems act with some independence, using tools and memory.
The guidelines add extra safeguards for them.