Follow us on Google News
India’s national cybersecurity agency has released a new framework for auditing artificial intelligence systems, asking organisations to assess risks involving autonomous AI agents, data leaks, malicious manipulation and inadequate human oversight.
The Indian Computer Emergency Response Team (CERT-In) published its 107-page Artificial Intelligence Security Auditing Guidelines on November 4, 2026. The framework covers AI systems developed, purchased, deployed or operated by organisations in India, as well as systems processing Indian data or providing AI-enabled services to Indian users.
The guidelines introduce four risk categories and establish procedures for evaluating AI security throughout a system’s lifecycle.
They also incorporate CERT-In’s existing six-hour cybersecurity incident-reporting requirement and specify how organisations should prepare for AI-related incidents.
The development comes as AI tools increasingly enter banking, healthcare, education, public administration and other sectors where inaccurate decisions or security failures can have serious consequences.
Four Risk Categories To Determine AI Audit Requirements
The framework classifies AI systems into four categories: low, moderate, high and critical risk.
An organisation must assess factors including the potential consequences of a failure, the sensitivity of data processed, the system’s independence in making decisions and the possibility of reversing harmful outcomes.
For example, an AI tool that performs a routine administrative function may present fewer risks than an automated system authorised to initiate financial transactions.
A system operating critical infrastructure could require more extensive safeguards if its failure threatens essential services or public safety.
The assigned category determines the depth of auditing, documentation, monitoring and human supervision expected.
Low-risk systems can undergo relatively simple internal assessments, while higher-risk applications face increasingly detailed examinations.
High-risk systems are expected to undergo independent third-party audits, with critical-risk systems subject to more extensive independent assessments and continuous monitoring.
The framework also provides for reassessment when an AI system’s capabilities, operating environment or potential impact change.
This approach recognises that the risks associated with an AI system may increase after deployment, particularly when additional functions or greater autonomy are introduced.
Six-Hour Reporting Rule Extended To AI Incident Audits
One of the most important provisions concerns the handling of cybersecurity incidents involving artificial intelligence.
CERT-In states that AI incidents meeting its reportable criteria must be notified within six hours.
However, this deadline is not a newly introduced reporting obligation.
India’s national cybersecurity agency established the six-hour requirement under directions issued on April 28, 2022, covering specified categories of cybersecurity incidents.
The November 2026 guidelines incorporate that existing obligation into a dedicated AI security-auditing framework.
This distinction matters because an ordinary AI error or inaccurate response does not automatically qualify as a reportable cybersecurity incident.
Organisations must assess whether the event falls within the applicable reporting categories.
The framework also addresses incidents involving compromised AI models, unauthorised access, attacks against datasets, malicious manipulation and failures that could threaten system security.
Organisations are expected to maintain procedures for identifying incidents, preserving evidence, containing attacks and restoring affected systems.
The guidelines also address the retention of security logs for 180 days, with synchronised timestamps across AI components to support investigations.
Log records can help investigators reconstruct when suspicious activity occurred, which systems were affected and whether attackers gained access to confidential information.
The 180-day retention requirement also builds on CERT-In’s existing cybersecurity directions rather than establishing an entirely new retention period.
AI Agents And Chatbots Face Dedicated Security Checks
The framework identifies nine core auditing areas, including governance, data quality, privacy, model risk management, fairness, transparency, cybersecurity, continuous monitoring, human oversight and third-party accountability.
It also establishes specific considerations for advanced AI systems.
These include generative AI, multimodal models, retrieval-augmented generation systems and autonomous AI agents.
An AI agent differs from a conventional chatbot because it may be authorised to use external tools, execute commands or perform tasks without requiring approval at every stage.
Such capabilities can become dangerous if attackers manipulate the system’s instructions or gain access to the tools it controls.
The guidelines therefore address prompt injection, where malicious instructions are designed to redirect an AI system from its intended task.
They also cover data poisoning, in which attackers manipulate training information or other data used by a model.
Additional concerns include sensitive information leakage, insecure software components, excessive AI permissions and unsafe communication between multiple AI agents.
Organisations are expected to examine these risks through security testing, monitoring and documented controls.
The guidelines emphasise that automated AI-generated audit findings cannot replace reliable evidence or independent human judgement.
This is especially important when AI tools are used to evaluate other AI systems, creating the possibility that inaccurate automated conclusions could go undetected.
Companies Must Examine AI Throughout Its Lifecycle
CERT-In’s approach extends security assessment beyond the moment an AI product is launched.
The framework covers the collection of training data, model development, testing, deployment, ongoing monitoring and eventual retirement of a system.
Organisations are expected to maintain records explaining the intended use of AI applications, their technical architecture, relevant risks and controls.
The guidelines also emphasise the traceability of training data, system configurations and significant changes made after deployment.
This could help auditors identify whether an AI system has been altered, compromised or used beyond its approved purpose.
Third-party AI providers form another area of scrutiny because organisations may depend on externally developed models, datasets and software.
The framework seeks to establish clearer accountability when these technologies are incorporated into business and government operations.
It does not, by itself, establish a new standalone AI law or make every provision an independently enforceable statutory obligation. Applicable legal requirements and the framework’s audit expectations must be distinguished.
For Indian organisations, the immediate significance is a more detailed basis for examining AI-related cybersecurity risks rather than relying exclusively on conventional software security assessments.
What This Means For You:
Businesses deploying AI should review access permissions, maintain incident-response procedures and determine which systems require independent security assessments. Users should be cautious about sharing sensitive information with AI services whose privacy and security controls are unclear.
