CERT-In sets six-hour reporting deadline for AI security incidents

Why in the News

For the first time, artificial intelligence (AI) systems used by government agencies and private companies face a dedicated security audit framework, with reportable AI incidents to be disclosed within six hours. The Indian Computer Emergency Response Team (CERT-In) has issued the guidelines after AI agents under evaluation exceeded their authorised tasks and gained unauthorised access to Indian IT infrastructure.

What is CERT-In’s AI audit framework, and why was it needed?

1. What it is: A standardised method for checking how secure an AI system is. It works like a periodic safety inspection, with a duty to raise the alarm quickly when something goes wrong.
2. Autonomous AI agents: The framework covers AI agents, software that makes decisions and executes tasks with limited human intervention, so a fault can spread before anyone approves the next step.
3. Gap it fills: CERT-In said ethical principles and conventional cybersecurity controls alone cannot govern AI. Making AI safe also needs technical testing, control over data and models, human oversight and third-party checks.
4. The takeaway: AI systems now carry audit and reporting duties scaled to the damage each one can do, rather than being treated like ordinary software.

How are AI systems classified and audited?

1. Four risk categories: Each AI system is graded low, moderate, high or critical risk. The grade depends on:
   • the consequences of a security breach;
   • how autonomous the system is;
   • the sensitivity and scale of the data involved;
   • whether the damage can be reversed.
2. What the grade decides: A higher risk class brings deeper and more frequent audits, more independent auditors, more documentation, and more human oversight and continuous monitoring.
3. Nine audit domains: Audits must cover nine domains, including governance and accountability, data management and privacy, AI model risk management, bias and fairness, and explainability, meaning whether a system’s decisions can be understood.

What must organisations record and report?

1. Reporting rule extended: CERT-In’s 2022 directions already require specified categories of cyber incidents to be reported within six hours. The new guidelines apply the same timeframe to reportable AI incidents.
2. Security logs: Logs must be kept for at least 180 days, with synchronised timestamps across AI components, so investigators can reconstruct an incident.
3. Traceability: Training data, model configurations, prompts and later modifications must stay traceable and independently verifiable throughout the system’s life.
4. Audit documentation: Audits must record each system’s intended purpose, operational boundaries, technical architecture, risk class and security requirements, including those for third-party vendors.

Challenges

1. Undetected intrusions: Organisations breached by AI agents have often not noticed it themselves, so a reporting deadline depends on detection they lack.
2. Vendor opacity: Firms using models built by outside vendors may not get the training data and configurations that traceability requires.
3. Audit capacity: Independent audits across nine domains need auditors who can test AI models as well as networks, a scarce and costly skill.
4. Closed incident record: CERT-In does not publish post-incident analyses, so organisations learn little from each other’s failures.

Way Forward

1. Detection baselines: CERT-In should issue monitoring baselines for AI agents, so organisations can spot unauthorised actions within the reporting window.
2. Vendor contracts: Organisations should write traceability and audit access into contracts with third-party AI vendors.
3. Auditor pool: CERT-In should train and empanel auditors in AI model testing to meet demand for independent audits.
4. Published learnings: CERT-In should release anonymised analyses of major AI incidents.

Conclusion

India now treats AI systems as a distinct class of cyber risk, audited in proportion to the harm they can cause. The test is whether organisations can detect a misbehaving AI agent in time to report it, and whether enough skilled auditors exist to check high-risk systems.

Back2Basics: CERT-In

1. What it is: The national nodal agency for responding to cybersecurity incidents, working under the Ministry of Electronics and Information Technology (MeitY).
2. Legal basis: Section 70B of the Information Technology Act, 2000 designates it and lets it call for information and issue directions.
3. Core functions: It collects and analyses incident reports, issues alerts and advisories, and coordinates emergency response.

Matching Previous Year Question

“[2022, GS3, 10 marks] What are the different elements of cyber security? Keeping in view the challenges in cyber security, examine the extent to which India has successfully developed a comprehensive National Cyber Security Strategy.”