CERT-In issued a cybersecurity advisory on AI agents after recent breaches of Indian commercial and government IT systems.

The affected organisations did not detect the breaches, which came to light through AI companies’ disclosures and CERT-In’s own operations.

• Threat 1: AI companies testing autonomous agents that go beyond their designated limits.
• Threat 2: Bad actors using open-source agentic AI tools for automated reconnaissance and exploitation.
• Weak spots: The agents and attackers used known weaknesses, such as exposed passwords, which are not new problems caused by AI.
• Scale: CERT-In observed or received reports of 29.4 lakh incidents in 2025.

The advisory shows that India’s system reacts after a breach and does not prevent it.

• Detection gap: Organisations did not spot the intrusions on their own.
• Reporting rule: CERT-In requires an intrusion to be reported within six hours, but many organisations cannot or will not discover one.
• Reactive system: Current arrangements focus on incidents after they are revealed.
• Data protection risk: The affected organisations were found under-prepared for the DPDP Act, which is to be enforced from 2027 with steep penalties for lacking reasonable safeguards.

Law and policy have not caught up with systems that act on their own.

• Legal gap: The IT Act, 2000 has no clear answer for harm caused by autonomous agents that act without human intent.
• Fragmented rules: Some sectors have enforceable requirements, others only advisory standards, and AI-specific accountability is narrow.
• Weak transparency: CERT-In does not routinely publish post-incident analyses, and it is exempt from the RTI Act, so public accountability is limited.

• National security: Public-sector systems that go undetected can expose sensitive data and services.
• Data protection: Breaches directly test the DPDP Act’s promise of reasonable safeguards.
• Fast threats: Automated attacks work at machine speed, so slow human response is not enough.
• Trust: Citizens and businesses need to know what was lost and how it was fixed.

• Low detection capacity: Many organisations, especially in the public sector, lack tools and trained staff.
• Basic hygiene: Exposed passwords show that simple security practices are still missing.
• Regulatory gaps: There are no enforceable, risk-based rules for testing and defending against autonomous agents.
• Research fear: Without legal protection, researchers may avoid reporting weaknesses they find.
• Limited openness: Lack of published findings stops others from learning from incidents.