Quick answer: The AI Security Auditing Guidelines are a 107-page framework from the Indian Computer Emergency Response Team (CERT-In) for checking whether AI systems are secure, fair and auditable. Version 1.0 is dated 4 November 2026. It sorts AI systems into four risk tiers, sets nine audit domains, and adds controls for generative and agentic AI.
Why CERT-In’s AI security auditing guidelines are in the news
CERT-In has posted the “Artificial Intelligence (AI) Security Auditing Guidelines” on its website. Subtitled “Framework for Secure, Ethical and Auditable Artificial Intelligence Systems”, it was prepared by Sub-group 03 of CERT-In’s working group on “Indian Standards for Cybersecurity Auditing”.
The preamble notes that AI is now used in banking, healthcare, telecom, defence and citizen services. It brings risks that normal IT audits may miss, such as poisoned training data, deepfakes and hallucinated outputs. CERT-In has also set up a dedicated AI cyber defence “War Room”.
Key facts about the AI Security Auditing Guidelines
- Who it covers: AI systems developed, bought, deployed or run by organisations in India, and also AI systems that process Indian data or serve Indian users.
- Four risk tiers: Low, Moderate, High and Critical. They are based on factors such as impact severity, autonomy, data sensitivity, scale, reversibility and how critical the sector is.
- Nine core audit domains cover the whole AI lifecycle, from design and procurement to retraining and decommissioning.
- Seven guiding principles, aligned with MeitY’s India AI Governance Guidelines (2025).
- Incident reporting: AI incidents that meet CERT-In’s reportable criteria must be reported within 6 hours, and security logs must be kept for 180 days.
- Agentic AI checks: agents should have least-privilege permissions, human approval gates and a kill switch. High-risk autonomous systems must be able to shut down safely or move to a safe state.
- Maturity: a five-level model runs from “Initial/Ad hoc” to “Optimising”.
- Human judgement: auditors may use AI tools, but AI outputs cannot replace the auditor’s independent judgement.
Risk tiers and what each demands
| Tier | Minimum audit requirement |
|---|---|
| Tier 1 (Low) | Self-assessment or basic internal review, reviewed annually |
| Tier 2 (Moderate) | Documented internal audit, plus an independent audit by a CERT-In empanelled organisation at least once a year |
| Tier 3 (High) | Periodic independent third-party audit by a CERT-In empanelled organisation, every six months where risk or major changes warrant |
| Tier 4 (Critical) | Independent audit by a CERT-In empanelled organisation with continuous monitoring and event-triggered reassessment |
A system can be moved up a tier if, for example, it controls critical infrastructure or cannot be reliably stopped or rolled back.
The nine domains and seven principles
| Nine core audit domains | Seven guiding principles |
|---|---|
| 1. AI governance and accountability 2. Data lifecycle, quality and privacy 3. Model risk management 4. AI ethics, bias and fairness 5. Explainability and transparency 6. Security, adversarial resilience and incident reporting 7. Continuous monitoring, drift management and validation 8. Human oversight and decision accountability 9. Third-party and AI supply-chain assurance | 1. Security-by-design 2. Accountability and human oversight 3. Transparency and explainability 4. Fairness and bias management 5. Data security and privacy 6. Auditability and traceability 7. Resilience and continuous assurance |
Extra requirements apply to Generative AI, Retrieval-Augmented Generation (RAG), agentic and multi-agent AI, covering threats such as prompt injection, data and model poisoning and unsafe tool use. The framework maps to ISO/IEC 42001, ISO/IEC 27001 and the NIST AI Risk Management Framework, among others.
Purely academic research AI not in production, and rule-based automation with no machine learning, are outside the mandatory scope.
Background: what CERT-In is
CERT-In describes itself as the national nodal agency for responding to computer security incidents, and it has been operational since January 2004. It works under the Ministry of Electronics and Information Technology (MeitY). The IT (Amendment) Act, 2008 gave it a statutory role, and it issues directions under Section 70B of the Information Technology Act, 2000. Its roles include 24×7 incident response and issuing security advisories.
Exam angle: how NDA, CDS and CAPF can ask this
- Cyber security bodies are a common science and technology topic. Know CERT-In, its parent ministry (MeitY) and the IT Act, 2000.
- Terms like agentic AI, RAG, prompt injection and deepfakes now appear in current affairs papers. Learn one-line definitions.
- AI security and critical infrastructure protection make a useful point for CAPF (AC) essays.
Practice MCQs
Q1. CERT-In functions under which ministry?
(a) Ministry of Home Affairs (b) Ministry of Defence (c) Ministry of Electronics and Information Technology (d) Ministry of Communications
Answer: (c) — CERT-In is under MeitY.
Q2. How many risk tiers do CERT-In’s AI Security Auditing Guidelines define?
(a) Two (b) Three (c) Four (d) Five
Answer: (c) — Low, Moderate, High and Critical.
Q3. Under the guidelines, reportable AI incidents must be reported to CERT-In within:
(a) 6 hours (b) 24 hours (c) 72 hours (d) 7 days
Answer: (a) — the guidelines keep the 6-hour reporting window.
Q4. ISO/IEC 42001, referred to in the guidelines, is a standard for:
(a) Quality management (b) AI management systems (c) Environmental management (d) Food safety
Answer: (b) — it covers AI management systems.
Frequently asked questions
What are CERT-In’s AI Security Auditing Guidelines?
They are a framework, version 1.0 dated 4 November 2026, for auditing the security, governance, privacy, fairness and transparency of AI systems used in India.
What is agentic AI?
AI systems that act on their own, such as running code or using tools, to reach a goal. The guidelines require least-privilege access, human approval gates and kill switches for them.
Do the guidelines apply to foreign AI companies?
The framework covers AI systems that process Indian data or offer AI services to users in India, as well as those run by organisations in India.
Who audits high-risk AI systems?
For High and Critical tiers, independent audits are to be done by CERT-In empanelled auditing organisations. Critical systems also need continuous monitoring.
Also read on NCA: Eka-1 and Dhir AI models under the IndiaAI Mission | Monthly defence current affairs PDFs | CAPF coaching in Chandigarh
Sources: CERT-In, AI Security Auditing Guidelines v1.0, CERT-In website. Facts as reported up to 8 November 2026.
Featured image background (representative): Carl Lender from Sunrise, USA / Wikimedia Commons, CC BY 2.0.
