Unknown vulnerabilities: on the cybersecurity advisory by CERT-In

The cybersecurity advisory by the Computer Emergency Response Team (CERT-In) has outlined two threat vectors vis-à-vis Indian commercial and government IT infrastructure: AI companies testing autonomous agents exceeding their designated operational constraints and bad actors weaponising open-source agentic AI tools for automated reconnaissance and exploitation. However, the organisations affected by recent breaches that the advisory was based on did not detect them: they came to light in the AI companies’ disclosures and CERT-In operations. Aside from developers allowing experimental systems to act outside their authorised bounds, the advisory highlights two problems for India — existing arrangements focus on cybersecurity incidents after they are revealed; and the implicated organisations being found under-prepared for the DPDP Act — to be enforced from 2027, including steep penalties on entities not maintaining reasonable safeguards against breaches. While CERT-In observed or received reports of 29.4 lakh incidents in 2025, it does not routinely publish postmortem analyses and is exempt from the RTI Act. As a result, public accountability is also weak: CERT-In found that the AI agents and/or bad actors exploited vulnerabilities, including exposed passwords, that are known security problems separate from AI agents’ abilities.

AI companies and labs have made several disclosures this year of agents unexpectedly breaching real-world infrastructure worldwide during testing. For India, however, the legal boundaries specified by the IT Act 2000 leave open an accountability gap for autonomous agents without the backing of human intentions. A more specific lacuna is that while CERT-In requires an organisation to report an intrusion within six hours, many organisations remain incapable of, or unwilling to, discover it. The reporting deadline itself has not ensured detection. Overall, the national cybersecurity landscape is fragmented across a pastiche of enforceable requirements in some domains, advisory technical standards in others, and narrow AI-specific accountability in yet others. As human and autonomous threat actors become more sophisticated, India needs to leapfrog to a landscape with enforceable risk-based requirements for both testing and defending against autonomous agents; better spotting unauthorised access, especially in the public sector; and a legal safe-harbour for security research conducted in good faith, in line with CERT-In policy. CERT-In should also publish more information, anonymised where necessary, which includes remediation outcomes and whether the affected individuals were informed. Finally, the government should also review the exemption for CERT-In from the RTI Act as such a sweeping exemption is unnecessary for effective cyber-defence.