CERT-In AI Security Auditing Guidelines: four risk tiers for chatbots, AI agents

India's national cyber security agency, CERT-In, on Wednesday, 4 November 2026, issued its first Artificial Intelligence (AI) Security Auditing Guidelines, a framework for checking whether AI systems used by organisations are secure, ethical and auditable. The 107-page document, version 1.0 and numbered CISG-2026-04, sorts AI systems into four risk tiers, Low, Moderate, High and Critical, and sets how often and how deeply each must be audited. It also lists extra checks for chatbots, AI agents and other systems that can be tricked through their inputs or misuse the tools they control. The guidelines came a day after the Reserve Bank of India (RBI) met bankers over a wave of AI-assisted cyber attacks and, according to a report, issued a confidential advisory asking non-banking finance companies (NBFCs) that use banking APIs to get security audits from CERT-In-empanelled auditors.

Until now, Indian companies deploying AI had general cyber security rules to follow, but no national checklist for the risks that are specific to AI. A chatbot can be manipulated by a cleverly worded message into revealing data it should keep private. An AI agent that can send emails or move money can be steered into doing the wrong thing. Retrieval-based systems can leak the documents they search. The guidelines name these risks and tell auditors what to test.

For start-ups, the risk tiers matter most. A small company selling an AI tool to a bank, an insurer or a hospital is likely to find its product placed in a higher tier, which means more frequent independent audits and more paperwork before a large client will sign. That raises costs, but it also gives start-ups a clear standard to build to and to show buyers. Auditing firms on CERT-In's panel gain a new line of business.

The timing ties the guidelines to real attacks. CERT-In had warned in an October advisory that attackers were stealing the API credentials of payment aggregators and partner banks to move money, possibly with AI tools. The RBI's reported advisory responds directly, by requiring audits of the API links between banks and lenders. One open question is how binding the AI guidelines are: one legal analysis notes that the document uses the language of minimum requirements but does not state its own legal enforceability.

Peepals calculation: Under the tiers as summarised, a High-risk AI system faces about two audits a year (12 months ÷ 6), twice as many as a Moderate one. For NBFCs already connected to bank APIs, the RBI's reported deadline of 3 February 2027 leaves 91 days from the guidelines' release on 4 November 2026 (30 + 31 + 30). Our calculation from the reported figures.

Short-term: Banks, NBFCs, fintechs and AI vendors will start mapping their AI systems to the four tiers and lining up empanelled auditors, especially for systems that touch money or personal data.

Long-term: If regulators such as the RBI and others adopt the tiers in their own rules, the guidelines could become the default standard for AI security in India, shaping how start-ups design and sell AI products.

Who is affected: AI start-ups and software vendors; banks, NBFCs and fintechs using AI or open APIs; CERT-In-empanelled audit firms; and customers whose data and money flow through AI systems.

What to watch next: The 3 February 2027 deadline reported for existing NBFC API clients to complete audits, and whether sector regulators formally adopt CERT-In's AI audit tiers.

Questions and Answers