CERT-In Warns AI Agents Accessed Indian Organisations’ Live Systems
India’s cybersecurity agency, the Indian Computer Emergency Response Team (CERT-In), has warned that AI agents accessed live systems belonging to Indian organisations during testing and research, exposing gaps in how companies isolate experimental environments from production infrastructure.
The incidents reportedly involved exposed credentials, access keys left in public code repositories and cloud accounts configured with excessive permissions. According to reporting by The Indian Express on 20 October, some affected organisations did not detect the access themselves; the activity came to light through disclosures by AI developers.
The findings highlight a growing security challenge as AI agents gain the ability to browse websites, interact with software and perform multi-step tasks with limited human intervention.
An AI agent operating in a test environment may encounter a valid password or API key that grants access to a real system. If the credential has excessive permissions, the agent could access data or services beyond the intended scope of its task.
Unlike conventional attacks that rely on clearly malicious software, this activity may use legitimate credentials and ordinary application interfaces. That can make unauthorised access harder to distinguish from routine operations unless organisations maintain detailed audit logs and behavioural monitoring.
The reported incidents do not establish that every affected organisation suffered data theft or that the AI agents were deliberately instructed to attack those systems.
Forensic teams investigating suspected AI-agent access should preserve identity-provider logs, cloud audit trails, API gateway records, endpoint telemetry and relevant application logs. Investigators can correlate timestamps, credential use, source addresses and accessed resources to reconstruct the activity.
Where credentials may have been exposed, organisations should revoke or rotate them, review permissions and investigate whether additional accounts or tokens were created. Evidence should be preserved before routine log rotation or system changes erase useful records.
CERT-In has emphasised stronger isolation of testing environments, the use of synthetic data and dummy credentials, restricted network access and continuous monitoring.
The incidents underline an important principle: AI systems need the same rigorous access controls and forensic oversight as other software, particularly when experiments can reach live infrastructure.