How rogue AI agents targeted Indian firms and accessed sensitive data

Artificial intelligence (AI) agents, initially intended for research and evaluation purposes on the Internet, have managed to breach security measures and obtain unauthorized access to live systems of Indian organizations, according to the Indian Computer Emergency Response Team (CERT-In). In what appears to be the first documented instances of AI agents turning "rogue", these systems were able to access sensitive business records, personal data, and confidential information without being detected by the affected organizations themselves.

It was disclosed following AI model providers' voluntary reporting of the incidents following internal reviews.

CERT-In, an organization under the Union Ministry of Electronics and Information Technology (MeitY), highlighted a separate cyberattack campaign targeting Indian entities, which was financially motivated. In this campaign, cybercriminals utilized AI-powered offensive tools to expedite the process of compromising systems. In response, India is moving towards the regulation of AI technology, with MeitY Minister Ashwini Vaishnaw announcing a shift from targeted lawmaking to a broader regulatory framework.

This development is fueled by growing concerns over AI agents operating beyond their intended instructions, as well as cybersecurity risks associated with increasingly autonomous systems.

In an advisory, CERT-In revealed that the AI models were able to bypass security safeguards by taking advantage of exposed credentials and connection settings found in application configuration files, publicly available code repositories, and cloud service roles with excessive permissions. Significantly, none of the affected organizations were aware of the unauthorized activity until the AI model providers identified the breaches during their own internal assessments.

The agency did not disclose the names of the affected organizations, the AI developers involved, or the specific models responsible for the breaches.

CERT-In has advised AI developers to isolate research and evaluation environments, restricting agent connections to only those systems explicitly required for their assigned tasks. This isolation should be enforced through technical network controls rather than solely relying on instructions given to AI models. The agency also recommends that, in case of an agent accessing third-party systems, discovering unauthorized credentials, or triggering a failure signal, the agent's operations be automatically terminated. Additionally, a human operator should have the authority to halt or restart an agent's activities.