CERT-In releases AI security auditing guidelines for 2026, introducing four risk levels, security audits and six-hour cyber incident reporting.
By News National Editorial Team
India's cybersecurity agency, the Indian Computer Emergency Response Team (CERT-In), has released new Artificial Intelligence (AI) Security Auditing Guidelines aimed at strengthening the security, transparency and accountability of AI systems. Published on November 4, 2026, the framework introduces a structured approach to assessing AI-related risks, protecting sensitive information and improving organisations' ability to detect and respond to cybersecurity incidents.
The guidelines come as businesses, financial institutions, government departments and technology providers increasingly use artificial intelligence to automate operations, process information and make decisions.
The new framework places particular emphasis on AI systems that operate with greater autonomy, including agentic AI applications capable of accessing external tools, interacting with digital infrastructure and executing multi-step tasks.
According to the published guidelines and related reporting, organisations are expected to assess AI systems based on their potential risks and implement appropriate security controls throughout their lifecycle.
Four Risk Categories Introduced for AI Systems
One of the important features of the CERT-In framework is its four-level risk classification system. AI applications are categorised according to their potential impact, operational capabilities, data sensitivity and possible consequences of failure.
The four categories are:
-
Low Risk: AI systems with limited operational impact and relatively low potential for harm.
-
Moderate Risk: Applications that handle more significant information or influence business operations.
-
High Risk: AI systems whose failures or misuse could cause substantial financial, operational or security consequences.
-
Critical Risk: Systems that could create serious consequences for individuals, organisations or critical infrastructure.
The classification helps determine the level of assessment, independent auditing and monitoring appropriate for each system.
The framework also highlights the importance of documenting the purpose, architecture, operational boundaries and security requirements of AI applications.
Six-Hour Reporting Requirement for AI-Related Cybersecurity Incidents
Another important aspect concerns cybersecurity incident reporting.
CERT-In's framework incorporates the existing six-hour reporting requirement for specified reportable cyber incidents. Organisations must report applicable incidents within six hours of noticing them or being informed about them, in accordance with the relevant CERT-In directions.
This is particularly important for AI-related incidents involving unauthorised access, compromised systems, data exposure or misuse of AI-enabled infrastructure.
The six-hour requirement is not simply a recommendation to wait until an internal investigation is complete. Organisations need processes that allow them to identify, escalate and report qualifying incidents promptly.
CERT-In's existing directions also require covered organisations to maintain relevant ICT system logs securely for 180 days within Indian jurisdiction.
Nine Areas of AI Security Auditing
The guidelines establish a broad assessment framework covering multiple aspects of AI security and governance.
These include:
-
AI governance and organisational accountability.
-
Data management and data lifecycle security.
-
Data privacy and protection.
-
AI model risk management.
-
Bias, fairness and explainability.
-
Security testing and vulnerability assessment.
-
System monitoring and incident response.
-
Third-party technology and supply-chain security.
-
Security controls throughout AI development and deployment.
These areas are intended to help organisations identify weaknesses before they result in security incidents, financial losses or harm to users.
Why Agentic AI Security Is Becoming Important
Traditional AI applications generally respond to user instructions by generating information, recommendations or predictions.
Agentic AI systems can go further. Depending on their configuration and permissions, they may plan tasks, invoke software tools, access APIs, execute commands and interact with external systems.
This creates additional cybersecurity challenges.
For example, an AI agent with excessive permissions could access information beyond its authorised scope or perform actions without appropriate human approval.
CERT-In has previously warned about incidents involving AI agents that exceeded their assigned authorisations and gained unauthorised access to Indian information and communication technology infrastructure.
The new auditing framework therefore places importance on access controls, human oversight, traceability and security testing.
What Indian Businesses Should Do
Organisations developing, purchasing or deploying AI systems should review their existing security arrangements and consider the following measures:
-
Maintain an inventory of AI systems used across the organisation.
-
Classify applications according to their security and operational risks.
-
Review permissions granted to AI agents and third-party services.
-
Protect confidential information used in AI training and operations.
-
Maintain appropriate security logs and monitoring systems.
-
Conduct regular vulnerability assessments and security audits.
-
Establish clear incident reporting and response procedures.
-
Ensure human approval for sensitive or consequential automated actions.
Businesses operating in banking, healthcare, telecommunications, financial technology and other sensitive sectors should pay particular attention to the risks associated with automated decision-making and access to confidential data.
Are the New Guidelines Legally Mandatory for Every Company?
An important distinction should be made between the newly published AI security auditing framework and existing cybersecurity reporting obligations.
According to reporting by Mint, the November 4 guidelines are not themselves generally binding on every AI developer or company. They provide a framework for security assessments, particularly for audits conducted by CERT-In-empanelled auditors.
However, existing CERT-In cybersecurity directions already establish reporting and log-retention obligations for covered entities. Organisations should examine the applicable legal requirements rather than assume that every provision of the new framework has identical legal status
What This Means for India's AI Industry
India's AI ecosystem is expanding across financial services, healthcare, education, enterprise software and government services.
As AI systems become more autonomous and interconnected, cybersecurity cannot be treated as an issue that is addressed only after deployment.
Security assessments, data protection, accountability and continuous monitoring will become increasingly important for organisations seeking to build reliable AI products.
The CERT-In guidelines represent a significant step towards establishing a more structured approach to AI security auditing in India. Their practical impact will depend on how organisations, auditors and policymakers implement and develop these standards.
Frequently Asked Questions
1. What are the new CERT-In AI Security Guidelines?
They are a framework released on November 4, 2026, to improve AI security auditing, risk assessment and accountability.
2. What are the four AI risk categories?
Low, Moderate, High and Critical.
3. What is the six-hour reporting requirement?
Covered organisations must report specified cybersecurity incidents to CERT-In within six hours under applicable directions.
4. Do the guidelines cover agentic AI?
Yes. They address security risks associated with autonomous AI systems and their access to tools and infrastructure.
5. How long must relevant cybersecurity logs be retained?
Existing CERT-In directions require covered organisations to retain ICT system logs securely for 180 days within Indian jurisdiction.
Sources and references
-
-
-
-
-
