E-Paper
Cert-In issues AI security audit guidelines, sets 4-tier risk framework
Cert-In emphasised that compliance requirements for high-risk and critical systems apply uniformly, with no exemptions or relaxed standards for smaller firms or startups.
Prefer HTon Google
The Indian Computer Emergency Response Team (Cert-In) has issued comprehensive security auditing guidelines for artificial intelligence (AI) systems that require organisations that have developed or deployed AI in India to conduct regular security assessments.
The 107-page document, titled AI Security Auditing Guidelines, also applies to organisations that handle Indian data, including foreign companies that offer AI services or tools to Indian users.
A working group at India’s national cybersecurity agency prepared the document, which was released on November 4.
The guidelines classify AI systems into four risk tiers - low, moderate, high and critical. - based on factors including autonomous decision-making capacity, data sensitivity, how many people it affects, and how easily any harm can be undone.
The risk level decides how stringent the audit is.
Low-risk systems can be checked through a self-assessment or a basic internal review. For moderate-risk systems, the document prescribes an annual independent audit by an auditing organisation empanelled with Cert-In. High-risk and critical systems also need independent audits by empanelled auditors, and critical systems are to be monitored continuously.
Cert-In emphasised that compliance requirements for high-risk and critical systems apply uniformly, with no exemptions or relaxed standards for smaller firms or startups.
The guidelines extend Cert-In’s existing rules, under which reportable incidents must be flagged within six hours and logs kept for 180 days, to AI-related incidents, and make compliance part of the audit.
Organisations that use generative AI must have a policy against making deepfakes of real people without their consent, and must mark AI-generated media with content provenance labels, such as C2PA or an equivalent.
Organisations running AI systems in production must maintain an ‘AI bill of materials’, a record of the models, datasets, prompts, tools and software each system relies on, scaled to its complexity and risk. Cert-In had earlier issued technical guidance on such records, the new document makes it an audit item.
AI systems that act on their own, known as agentic AI, face extra checks for risks such as prompt injection and misuse of tools, while critical systems must go through red-team attack exercises every year. If auditors find two or more critical problems, or give an unsatisfactory opinion, a fresh audit is required within three months.
To be sure, the document does not say under which legal provision the guidelines were issued, when they apply, or what happens to organisations that do not follow them. It names regulators such as the Reserve Bank of India, Securities and Exchange Board of India, Insurance Regulatory and Development Authority of India (IRDAI) and the Telecom Regulatory Authority of India as intended users, but sets no separate rules for these sectors.
