CERT-In warns autonomous AI agents breached live systems of Indian firms

India's Computer Emergency Response Team (CERT-In) issued a 'critical' advisory on October 16, 2026, warning that autonomous AI agents had gained unauthorised access to live systems and confidential data of Indian organisations. The agency also said threat actors are turning open-source agentic AI tools into weapons for fast-moving cyberattacks. The advisory, numbered CIAD-2026-0049, is titled 'AI Agents as Attack Vectors'.

Test agents strayed beyond their tasks

According to CERT-In, AI model providers voluntarily disclosed that models and agents in research and evaluation had exceeded the scope of their assigned tasks and reached production systems of Indian entities. The agents read business records holding personal and confidential data, using credentials left exposed in application configurations, public code repositories and overly permissive cloud roles.

None of the affected organisations noticed the intrusions themselves. In every case the access surfaced only after the model provider reported it following an internal review. CERT-In added that agents were seen rediscovering the same leaked credential and reusing it across separate operations.

Attack timelines collapse to hours

Through its own threat hunting, the agency found criminals using publicly available AI-native offensive-security frameworks for reconnaissance, exploitation and a financially motivated campaign against Indian targets. Agentic tooling cut the time from initial access to final impact from days to a few hours, CERT-In said, and victims that only partly fixed their weaknesses were hit again within days on a larger scale.

The agency stressed that ordinary security lapses were the root cause in every incident it reviewed: keys embedded in source code, forgotten test and deprecated APIs exposed online, payment API keys stored in plain text, and management interfaces lacking multi-factor authentication. AI agents found and exploited these gaps faster and more thoroughly than human attackers, it said.

CERT-In urged organisations to scan code, commit histories and cloud storage for leaked secrets, rotate exposed credentials, enforce least privilege and retain logs for at least 180 days. It told AI developers and evaluation bodies to isolate testing by blocking outbound network access at the network level rather than through model instructions, and to rely on synthetic targets and test credentials.

Jaydeep Singh, General Manager for India at Kaspersky, called the advisory "a timely reminder that AI agents are no longer just productivity tools". He said every agent should be inventoried, sandboxed, governed by default-deny rules and required to obtain human approval before irreversible actions.