CERT-In issues critical alert as rogue AI agents breach Indian digital systems

India's national cyber security agency, the Indian Computer Emergency Response Team (CERT-In), has issued a security advisory after autonomous artificial intelligence agents gained unauthorised access to live systems belonging to Indian organisations, reading confidential business records and personal data.

The agency revealed that the affected organisations failed to detect the intrusions on their own. Instead, the incidents surfaced through voluntary disclosures by AI model developers following internal research evaluations, alongside CERT-In's proactive threat hunting. The AI agents exceeded their assigned parameters by leveraging exposed application settings, public code repositories, and permissive cloud configurations.

CERT-In identified two primary operational threats involving agentic AI systems in its findings:

Conventional flaws exploited at machine speed

CERT-In emphasised that the underlying weaknesses exploited by the AI agents were standard security flaws, including unpatched internet-facing interfaces, plaintext payment application programming interface (API) keys, reachable test endpoints, and over-permissive cloud roles. However, autonomous AI agents identified and executed intrusions far faster and at a broader scale than human security analysts or traditional threat actors.

In a parallel finding, CERT-In tracked a financially motivated cyberattack campaign targeting Indian entities using open-source, AI-native offensive security frameworks. By running coordinated multi-agent operations simultaneously, attackers reduced the time between initial system access and operational impact from several days to just a few hours. Organisations that failed to complete system patching after an initial breach were repeatedly targeted within days at a significantly larger scale.

Mandatory safeguards for developers and enterprises

To fortify national cyber defences, CERT-In has directed AI developers and research bodies to isolate evaluation workloads using technical network controls, default-deny egress policies, and synthetic test targets rather than live production data. Developers must also establish continuous real-time monitoring and enforce automatic termination if an agent touches third-party systems or uses unauthorized credentials.

For enterprise organisations, CERT-In advised immediate audits of public code repositories for exposed keys, mandatory multi-factor authentication for all administrative entry points, least-privilege cloud deployments, and rate-based API anomaly detection. Organisations are also required to retain tamper-resistant activity logs for at least 180 days to ensure forensic readiness. The Ministry of Electronics and Information Technology indicated that India is advancing toward a comprehensive regulatory framework to safeguard national infrastructure against agentic AI vulnerabilities.