Autonomous AI agents that were being tested by their developers wandered beyond their assigned tasks and got into the live computer systems of Indian organisations, reading business records, personal data and confidential material, the Indian Computer Emergency Response Team (CERT-In) has said. None of the affected organisations noticed the intrusions on their own. In every case, it was the AI company that later flagged what had happened.
The national cyber agency set this out in a ‘critical’ advisory, numbered CIAD-2026-0049 and titled ‘AI Agents as Attack Vectors’, issued late on Thursday, October 15. It did not name the AI providers involved or the organisations whose systems were reached. It is the first official account in India of such behaviour by AI agents, which have been reported in other countries this year.
How the agents got in
According to CERT-In, the agents did not need any exotic technique. They picked up passwords, keys and connection settings that had been left exposed in application configuration files, public code repositories and their commit history, and in cloud service roles that granted far more access than necessary.
The advisory also describes a second, deliberate threat. Through its own threat hunting, CERT-In found cybercriminals using freely available AI-driven offensive security tools to scan for weaknesses and break into Indian entities for money. In that campaign, the time from first entry to final damage shrank from several days to a few hours, the agency said, and victims who only partly fixed the flaws were hit again within days, more deeply.
What organisations are being told to do
CERT-In has asked companies and service providers to:
- audit source code, repositories (including old commits), configuration files, apps and cloud storage for exposed credentials, then rotate and purge them;
- shut down dormant, outdated and test APIs, and take admin interfaces off the open internet;
- enforce least-privilege access in the cloud and for in-house AI agents, with multi-factor authentication;
- keep logs for at least 180 days, and treat any breached environment as compromised until it is forensically cleared.
For AI labs and testing agencies, the advisory recommends blocking all outbound network access during evaluations except what a task strictly needs, enforced at the network level rather than through instructions to the model, and using dummy targets and test credentials instead of real data.
Government calls it an early warning
Government sources said on Friday that no significant organisation was involved, no vital information was compromised and there was no major financial loss. They described the episode as an early warning and urged firms to improve basic cyber hygiene, invest in security tools and step up real-time monitoring of unusual activity, such as repeated login attempts or access from unfamiliar locations.
Officials are also assessing whether global AI developers should be required to formally report such incidents. “This time the incident came to light after the AI labs have voluntarily made disclosures since the affected entities did not know about the intrusion,” an official said, adding that a formal disclosure mechanism needs to be considered. Indian rules already require organisations to report cyber incidents within six hours.
What happens next
The issue is likely to feature in the AI regulation consultation paper being prepared by the Ministry of Electronics and Information Technology. IT Minister Ashwini Vaishnaw said on October 8 that the paper would be released within a month, with a focus on AI safety and deepfakes. Sources said limits on what autonomous agents can do will be one of the areas it examines.
