India is planning tighter requirements for reporting artificial intelligence-related incidents, including clearer rules on what information companies must provide and how quickly they must report such events, as the government works to align existing cybersecurity obligations with its broader AI governance framework.
What India Is Actually Planning to Change
Sources within the Ministry of Electronics and Information Technology (MeitY) indicate the government intends to tighten the norms, timeframe, and content of AI-related incident reporting. The issue has gained real urgency as AI systems increasingly operate with greater autonomy, and the proposed approach could cover incidents involving AI and machine learning systems more broadly, including cases where AI agents act beyond their intended tasks entirely.
India's 2025 AI Governance Guidelines already propose a broader mechanism for tracking harms caused by AI. The emerging question now is how that framework will actually work alongside the country's existing mandatory cybersecurity incident-reporting regime, two systems that currently overlap in some areas but aren't yet fully integrated.
What Companies Must Already Report
Under CERT-In's 2022 Cyber Security Directions, service providers, intermediaries, data centres, body corporates, and government organisations are required to report specified cyber incidents within six hours of noticing them or being informed about them.
This mandatory list includes data breaches, data leaks, unauthorised access, attacks on cloud systems, and other attacks, and importantly, it already covers malicious or suspicious activity affecting systems, software, or applications related to AI and machine learning. CERT-In guidance describes attacks targeting machine learning models as including attempts to cause malfunction or improper behaviour, potentially using techniques like deceptive data or environment manipulation to corrupt a model and cause it to behave incorrectly.
The existing CERT-In framework, in other words, already covers cyberattacks specifically targeting AI systems. What remains genuinely unclear is whether MeitY's planned changes will tighten reporting requirements specifically for AI-related cybersecurity incidents, or whether future rules might extend to cover AI systems that cause harm or behave unexpectedly even without an underlying cyberattack driving the failure.
What Could Actually Count as an "AI Incident"
India's 2025 AI Governance Guidelines take a considerably broader approach than the existing CERT-In cybersecurity framework. They adopt the OECD's definition of an AI incident, an event or series of events in which the development, use, or malfunction of an AI system directly or indirectly causes harm. That harm can include harm to health, disruption of critical infrastructure, human-rights violations, or damage to property, communities, or the environment.
The guidelines separately identify a wider set of risks: malicious use of AI, bias and discrimination, transparency failures, systemic risks, loss of control over AI systems, and national-security threats. This means the proposed AI governance approach genuinely extends well beyond conventional cyberattacks. However, it's important to note these guidelines are recommendations for India's AI governance framework, they don't themselves replace CERT-In's mandatory cyber incident-reporting rules; the two currently exist as parallel, related but distinct systems.
Who Might Actually Be Required to Report
Under the existing CERT-In Directions, reporting obligations apply to service providers, intermediaries, data centres, body corporates, and government organisations, individual citizens aren't covered by these directions at all.
AI systems, though, often involve multiple organisations working together: one company may develop the underlying model, another provides access to it, and a third deploys it within an actual product or service. The AI Governance Guidelines recommend clear accountability across this entire AI value chain, with liability linked to the specific function each party performed, taking into account risk level and whether an entity exercised proper due diligence.
Notably, the guidelines don't currently create a new mandatory AI incident-reporting obligation specifying exactly whether the developer, provider, or deployer must be the one to actually make the report. Sorting out responsibility across this value chain is therefore one of the genuine challenges MeitY would need to address if it decides to expand the reporting framework meaningfully.
How Quickly AI Incidents Might Need to Be Reported
For cyber incidents already covered under CERT-In's existing directions, the current reporting deadline is six hours from when an organisation notices the incident or is informed about it. Companies don't need a complete investigation finished before making that initial report, CERT-In guidance allows entities to share whatever information is available at reporting time and submit additional details later within a reasonable period.
MeitY's reported plan to tighten this timeframe doesn't yet specify whether the existing six-hour requirement itself will change, or whether a separate deadline might apply specifically to certain AI incidents. That distinction could matter quite a bit in practice: some AI failures may genuinely require examining system inputs, outputs, model behaviour, and autonomous actions to establish what actually happened, a more involved process than simply identifying a straightforward cyberattack.
Expert Perspective
Prof. Triveni Singh, renowned cybercrime expert and former IPS officer, said: "AI incidents can move much faster than traditional cyber incidents. An autonomous system may take actions before a human even understands what has happened. Companies should know exactly what must be reported, who is responsible and how quickly they must act. Clear reporting rules will help India identify AI risks early and prevent the same failures from spreading."
What Information Companies Might Have to Disclose
MeitY is also considering tighter requirements around the actual content of incident reports, though the precise additional information companies might need to provide isn't yet specified. The existing CERT-In regime already imposes substantial information-retention requirements: covered organisations must maintain ICT system logs for a rolling 180 days, provide those logs to CERT-In when reporting an incident or when directed to, and CERT-In can seek additional information in a specified format and timeframe as needed.
The OECD's AI-specific framework offers one possible reference point for more detailed reporting. It sets out 29 criteria across eight areas, covering basic incident information, severity and type of harm, affected people, economic context, training data and inputs, information about the AI model itself, the system's task and level of autonomy, and measures taken to stop, prevent, or mitigate the harm, with seven of these criteria being mandatory under the OECD's own framework. There's no indication India has actually decided to adopt these specific 29 criteria; the connection here is simply that India's AI Governance Guidelines use the same OECD definition of an AI incident while the government continues developing its own proposed reporting framework.
How This Might Fit Alongside CERT-In
India is also planning a national AI incident database, separate from CERT-In's existing mandatory cybersecurity reporting system. The 2025 AI Governance Guidelines recommend creating a national database specifically to collect evidence about real-world harms caused by AI, a central system capable of pulling information from smaller databases maintained by authorised entities or sectoral regulators under a common reporting structure. The guidelines also recommend using existing mechanisms, like CERT-In's system, to monitor AI vulnerabilities in critical sectors.
Participation in this proposed AI incident database is envisioned quite differently from CERT-In's mandatory cybersecurity reporting requirements: organisations would be encouraged to report incidents through protocols that protect confidentiality, while the database would also draw information from other sources, including media reports and publicly available research, rather than relying solely on mandatory organisational disclosure.
The government's overall approach, then, currently rests on two distinct elements: CERT-In's mandatory reporting regime for specified cybersecurity incidents (including attacks and malicious activity affecting AI and machine learning systems), and a broader AI governance framework designed to collect information about real-world AI harms more generally. How these two systems will eventually work together, and whether mandatory CERT-In requirements will actually be expanded specifically for AI incidents, remains to be seen as MeitY continues developing this framework.
FAQs
Q1. What is the current reporting deadline for cyber incidents under CERT-In?
Six hours from when an organisation notices the incident or is informed about it, with additional details permitted to follow within a reasonable period afterward.
Q2. Does the existing CERT-In framework already cover AI-related incidents?
Yes, but only cyberattacks targeting AI and machine learning systems. It's unclear whether new rules will extend to AI systems causing harm without an underlying cyberattack.
Q3. Who is currently required to report incidents under CERT-In's directions?
Service providers, intermediaries, data centres, body corporates, and government organisations, individual citizens aren't covered.
Q4. What is India's proposed national AI incident database?
A separate system from CERT-In, recommended under the 2025 AI Governance Guidelines, designed to collect evidence about real-world AI harms from authorised entities, sectoral regulators, and public sources like media reports.