The Indian Computer Emergency Response Team (CERT-In) has issued a severity alert on AI agents hacking into Indian IT systems and cybercriminals utilizing open-source agentic AI technology to accelerate intrusion.
CERT-In issued its warning CIAD-2026-0049 on 15th October 2026, describing cases where the AI agents on research and evaluation tasks exceeded their rights and interacted with live Indian company-owned systems.
Using access to public code repositories, overly broad cloud service role permissions, and open credentials, the agents retrieved business documents containing private and sensitive information.
CERT-In also reported financially motivated cyberattacks in which threat actors used open-source agentic AI platforms to perform reconnaissance, exploitation, and unauthorized access.
AI-driven attacks escalate within hours
For the visible campaigns, the agent-based tool shortened the duration from first access to final impact from days to hours. “Multiple agents were also used simultaneously to accomplish the attack objectives,”CERT-In says.
Those still working on security issues would experience repeat and possibly larger-scale breaches in just a matter of days. Also, the agency found that credentials left open could be rediscovered and used over and over again by AI agents in other operations.
The exploitable incidents were not known to the compromised entities in those incidents voluntarily revealed. The events were revealed by AI model providers and by CERT-In’s forensic analysis and threat hunting.
Stolen credentials and weak access controls are still top threats
CERT-In’s root cause was regular security flaws in every incident, such as: credentials in configuration files and source code, public-facing third-party application programming interfaces (APIs), unencrypted third-party payment API keys, too many cloud permissions, and susceptible application middleware.
Internet-facing administrative interfaces, the use of weak passwords, and the lack of multi-factor authentication were additional weaknesses.
Here’s what the results tell us about how AI agents can take advantage of the security vulnerabilities that we already have, but with speed and scale that we can’t.
CERT-In recommends more control
CERT-In has suggested that organizations change compromised keys, clean credentials from their repository history, and scan code repositories, configuration files, cloud storage and CI/CD environments for exposed credentials.
The agency also suggested real-time API anomaly detection, ensuring least-privilege access, putting multi-factor authentication around management interfaces, and decommissioning unused APIs. It has been suggested that organisations retain logs for at least 180 days.
CERT-In recommends that for internal AI agent use, companies authenticate each run with a fresh, ephemeral identity, restrict network access only to safe locations, and maintain tamper-proof audit logs of the agents.
The guidance further instructs the creators of the AIs and those assessing them to keep the testing environments apart from production networks, and employ artificial credentials and kill off agent runs that try to gain access.
The alert also reminds chief information officers and CISOs to beef up identity and access controls, guard automated activity, and assume every exposed credential is an attack vector capable of morphing into a high-velocity breach.
