CERT-In issues AI Security Auditing Guidelines v1.0 with four risk tiers and independent audit for high-risk AI

In short

CERT-In has released version 1.0 of its AI Security Auditing Guidelines, a 107-page framework for auditing the security, governance and accountability of AI systems. AI systems are sorted into four risk tiers; the higher the tier, the deeper and more independent the audit, with the top two tiers needing audit by a CERT-In empanelled auditing organisation. The document is dated 4 November 2026.

Authority

Indian Computer Emergency Response Team (CERT-In), Sub-group 03 of its Working Group on Indian Standards for Cybersecurity Auditing

Document

AI Security Auditing Guidelines v1.0, dated 4 November 2026

Type

Guidelines (audit framework)

Risk tiers

Tier 1 Low, Tier 2 Moderate, Tier 3 High, Tier 4 Critical (composite score 8 to 32)

Audit domains

Nine, from AI governance to third-party and supply-chain assurance

Maturity model

Five levels

Aligned with

MeitY India AI Governance Guidelines, ISO/IEC 42001, 23894 and 27001, NIST AI RMF, MITRE ATLAS, OWASP guidance

What changed

The framework covers AI developed, bought or used through APIs, and also generative, agentic and multi-agent AI. Research AI that is not deployed to the public is outside the mandatory scope. The document marks some requirements as mandatory for a tier and others as recommended; a referenced standard becomes mandatory only if the Guidelines or another applicable external requirement make it so. The text read does not name a commencement date, a transition period or the statutory provision under which it is issued, and it does not say which regulators will make it binding on their regulated entities.

Who it applies to

• Organisations that build, buy or run AI systems: government bodies, public sector entities, critical infrastructure operators, regulated entities and private companies
• CERT-In empanelled cybersecurity auditing organisations
• Internal and external auditors, chartered accountants in assurance and risk roles, and compliance and risk officers
• AI vendors, SaaS and cloud providers and data labelling or outsourced AI development firms in the supply chain

What to do

1. List the AI systems in use and score each on the risk factors to arrive at a tier; repeat at least annually.
2. For Tier 3 and Tier 4 systems, plan independent audit by a CERT-In empanelled auditor.
3. Prepare evidence across the nine domains: governance, data and privacy, model risk, bias, explainability, security, monitoring, human oversight and third-party assurance.
4. Check whether a sector regulator or contract already makes the framework binding for you before treating it as mandatory.

Reference

CERT-In, AI Security Auditing Guidelines v1.0, 4 November 2026.

Written by TaxByKK from the official notification. Check the official text before you act on it.

Related