CERT-In Writes India's AI Security Rulebook: Four Tiers, Six Hours
AI PolicyNovember 05, 20264 min readBy Digitalaikart Editorial

CERT-In Writes India's AI Security Rulebook: Four Tiers, Six Hours

#AIPolicy#AISafety#CERTIn#IndiaAI

TL;DR

01.A rulebook arrives before the law does

On November 4, 2026, the Indian Computer Emergency Response Team (CERT-In) published a 107-page document titled Artificial Intelligence Security Auditing Guidelines: Framework for Secure, Ethical, and Auditable Artificial Intelligence Systems. It was prepared by a working group inside the national cybersecurity agency and released on a Wednesday, according to reports in the Hindustan Times and Mint.

The scope is wider than most Indian AI guidance so far. The framework covers AI systems developed, purchased, deployed or operated by organisations in India, and also systems that process Indian data or offer AI-enabled services to Indian users. That second clause pulls in foreign companies running AI products in the country, including the big chatbot platforms. Mint noted one important nuance: the guidelines are not themselves binding on companies building AI. They are legally binding only on cybersecurity auditors empanelled with CERT-In, which makes them the yardstick against which AI systems inside companies will be audited.

02.Four tiers, four levels of scrutiny

The core of the framework is a classification system that places AI models and applications into four tiers: low, moderate, high and critical risk. The tier is decided by the potential consequences of a security breach, the autonomy of the system, the sensitivity and scale of the data involved, and whether the damage caused can be reversed or recovered.

The tier then decides what an audit looks like. Low-risk systems can be cleared through a self-assessment or a basic internal review. Moderate-risk systems need an annual independent audit by an organisation empanelled with CERT-In. High-risk and critical systems also require independent audits by empanelled auditors, with critical systems monitored continuously. CERT-In has been explicit that the requirements for high-risk and critical systems apply uniformly, with no relaxed standards for smaller firms or startups, the Hindustan Times reported.

03.Nine areas that every audit must cover

Whatever the tier, organisations have been asked to conduct security audits across nine domains. Business Standard listed them as including governance and accountability, the data lifecycle, data quality and data privacy, AI model risk management, bias and fairness, and AI system explainability, alongside monitoring and incident-response expectations.

Two obligations reach into everyday operations. Reportable AI-related security incidents must be flagged to CERT-In within six hours, extending the six-hour rule that has applied to specified cyber incidents since CERT-In's 2022 directions. And organisations must keep logs of their AI systems' usage and operations for 180 days, so that an incident can be reconstructed after the fact. For a mid-sized company running a customer-support bot or a credit-scoring model, that means logging and incident-response processes have to be designed in from the start, not bolted on later.

04.Why AI agents get their own tests

The guidelines single out agentic AI, systems that can plan and carry out multi-step tasks, use external tools, execute code and delegate work to other AI systems. The Hindu reported that autonomous task execution, persistent memory and the ability to delegate work are identified as factors that raise risk, particularly when they operate without human intervention.

The framework calls for testing whether an AI agent can exceed its authorised permissions, use tools without approval, bypass approval mechanisms, or misuse credentials. It also asks auditors to examine unsafe delegation of tasks between agents, the effectiveness of emergency-stop mechanisms, and whether multi-agent systems keep authenticated, logged communications and records of their actions. In plain terms, the questions are the ones an IT team would ask of a new employee with system access: what can this thing touch, who said it could, and can we switch it off?

05.What it means for Indian builders, and what comes next

The timing is not accidental. On November 3 and 4, CERT-In and the Indian Cyber Crime Coordination Centre issued advisories warning banks, NBFCs and fintech firms about a campaign of AI-driven attacks that led to unauthorised fund transfers through mule accounts and IMPS and NEFT transfers, the Hindustan Times reported. Auditable logs and tested agent permissions are the practical answer to that class of incident.

For Indian startups and enterprises, the useful move now is mapping: put each AI system you run into one of the four tiers, write down the data it touches and the tools it can call, and check whether your logs would survive a 180-day lookback. Companies building agents should test permission boundaries and kill switches before an external auditor does. The open questions are real, though. The document does not spell out which legal provision it is issued under, exactly when obligations begin, or what follows for organisations that ignore it, and Mint notes that consequences may only arrive once India's AI regulations take shape. Until then, the guidelines are the audit standard rather than a penalty regime.

Frequently asked questions

What are CERT-In's AI Security Auditing Guidelines?
A 107-page framework published on November 4, 2026 that sets out how AI systems used in India should be classified by risk, audited, logged and monitored, and how AI-related security incidents should be reported.

Which AI systems do the guidelines cover?
AI systems developed, purchased, deployed or operated by organisations in India, plus systems that process Indian data or provide AI-enabled services to Indian users, which includes foreign AI providers.

Are the guidelines legally binding on companies?
Not directly. They are legally binding only on CERT-In-empanelled cybersecurity auditors, and serve as the basis on which AI systems inside companies will be audited. Consequences for non-compliance may come with India's future AI regulations.

Found this useful? Share it

Source