The Indian Computer Emergency Response Team (CERT-In) has introduced a dedicated set of guidelines to help organisations assess and manage security threats linked to AI deployment. As artificial intelligence becomes part of critical business and public services, India’s cybersecurity framework is expanding to address the risks associated with increasingly autonomous and data-driven systems.
Published on November 4, 2026, CERT-In’s 107-page Artificial Intelligence Security Auditing Guidelines cover AI systems developed, purchased, deployed or operated by organisations in India. The framework also applies to systems that process Indian data or provide AI-enabled services to users in the country.
The guidelines introduce 4 risk categories and outline procedures for evaluating AI security across a system’s lifecycle. They also incorporate existing cybersecurity incident-reporting requirements and provide directions for preparing for AI-related security incidents.
The framework comes amid growing AI adoption across banking, healthcare, education, public administration and other sectors where inaccurate decisions or security failures could have significant consequences.
Four risk categories determine audit requirements
CERT-In classifies AI systems into 4 categories: low, moderate, high and critical risk. Organisations must evaluate factors such as the potential consequences of system failure, the sensitivity of data being processed, the level of autonomy in decision-making and whether harmful outcomes can be reversed.
A routine administrative AI tool may present relatively limited risks compared with an automated system authorised to initiate financial transactions. Similarly, an AI system supporting critical infrastructure may require stronger safeguards if its failure could disrupt essential services or threaten public safety.
The assigned category determines the level of auditing, documentation, monitoring and human supervision required. Low-risk systems may undergo simpler internal assessments, while high-risk applications are expected to undergo independent third-party audits. Critical-risk systems face more extensive independent assessments and continuous monitoring.
Organisations must also reassess systems when their capabilities, operating environments or potential impact change, particularly when new functions or greater autonomy are introduced.
Existing six-hour incident reporting requirement included
One of the key provisions addresses cybersecurity incidents involving AI systems. CERT-In states that reportable AI-related incidents must be notified within 6 hours.
However, this is not a newly introduced reporting requirement. The six-hour deadline was established under CERT-In’s directions issued on April 28, 2022, which cover specified categories of cybersecurity incidents. The November 2026 guidelines incorporate this existing obligation into a dedicated AI security auditing framework.
An ordinary AI error or inaccurate response does not automatically qualify as a reportable cybersecurity incident. Organisations must determine whether an event falls within the applicable reporting categories.
The framework addresses incidents involving compromised AI models, unauthorised access, attacks on datasets, malicious manipulation and security failures. Organisations are expected to establish procedures for identifying incidents, preserving evidence, containing attacks and restoring affected systems.
It also addresses the retention of security logs for 180 days, with synchronised timestamps across AI components to support investigations. These records can help establish when suspicious activity occurred, which systems were affected and whether confidential information was accessed.
The 180-day retention period also builds on existing CERT-In cybersecurity directions rather than introducing an entirely new requirement.
Nine core areas and emerging AI threats under scrutiny
The guidelines identify key auditing areas covering governance, data quality, privacy, model risk management, fairness, transparency, cybersecurity, continuous monitoring, human oversight and third-party accountability.
They also address security considerations for advanced AI technologies, including generative AI, multimodal models, retrieval-augmented generation systems and autonomous AI agents.
Unlike conventional chatbots, AI agents may use external tools, execute commands and perform tasks without requiring approval at every stage. These capabilities can create security risks if attackers manipulate their instructions or gain access to the tools they control.
The framework highlights several threats, including:
- Prompt injection: Malicious instructions designed to redirect an AI system from its intended task.
- Data poisoning: Manipulation of training information or other data used by an AI model.
- Sensitive information leakage: Unauthorised exposure of confidential or personal data.
- Insecure software components: Vulnerabilities in the technologies supporting AI systems.
- Excessive permissions: AI systems being granted more access or authority than necessary.
- Unsafe agent communication: Security risks arising from interactions between multiple AI agents.
Organisations are expected to address these threats through security testing, continuous monitoring and documented controls.
The guidelines also emphasise that automated AI-generated audit findings cannot replace reliable evidence or independent human judgement. This is particularly relevant when AI tools are used to assess other AI systems, as inaccurate automated conclusions could otherwise go undetected.
Security assessments to cover the entire AI lifecycle
CERT-In’s framework extends security assessments beyond the initial deployment of an AI product. It covers training data collection, model development, testing, deployment, ongoing monitoring and eventual retirement.
Organisations are expected to maintain documentation outlining the intended use of AI applications, their technical architecture, identified risks and security controls. The guidelines also stress the importance of tracking training data, system configurations and significant post-deployment changes.
Third-party AI providers are another area of focus, as organisations increasingly rely on externally developed models, datasets and software. The framework aims to establish clearer accountability when such technologies are integrated into business and government operations.
The guidelines do not establish a new standalone AI law or make every provision an independently enforceable statutory obligation. Organisations must distinguish between existing legal requirements and the framework’s audit expectations.
For Indian businesses, the guidelines provide a more detailed basis for evaluating AI-related cybersecurity risks beyond conventional software security assessments. Organisations deploying AI systems should review access permissions, strengthen incident-response procedures and identify applications that require independent security audits.
Users should also exercise caution when sharing sensitive information with AI services whose privacy and security controls are unclear.
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream LinkedIn | The Mainstream Facebook | The Mainstream Youtube | The Mainstream Twitter
About us:
The Mainstream is a premier platform delivering the latest updates and informed perspectives across the technology business and cyber landscape. Built on research-driven, thought leadership and original intellectual property, The Mainstream also curates summits & conferences that convene decision makers to explore how technology reshapes industries and leadership. With a growing presence in India and globally across the Middle East, Africa, ASEAN, the USA, the UK and Australia, The Mainstream carries a vision to bring the latest happenings and insights to 8.2 billion people and to place technology at the centre of conversation for leaders navigating the future.