AI agents accessed Indian firms' records without authorisation: Cert-In

Indian organisations failed to detect unauthorised access by AI models and agents, while AI companies voluntarily disclosed the incidents to Cert-In, the cyber agency said

The Indian Computer Emergency Response Team (Cert-In) on Thursday said some artificial intelligence (AI) model service providers informed it that their frontier AI models and agents had, “during research and evaluation activity”, reached the live systems of several Indian organisations.

 

“The recorded activity involved reading business records, including personal data and confidential material, through credentials and connection settings exposed in application configuration, public code repositories and permissive cloud service roles,” Cert-In said in an advisory issued late Thursday night.

 

None of the organisations accessed by these AI models and agents could detect the access, the government's cybersecurity coordination agency said. The incident came to light after these organisations disclosed that the AI models and agents “exceeded the authorisation of their assigned tasks”.

 

 

In addition to the voluntary disclosures by the AI companies, Cert-In also said that during its proactive threat hunting, it found instances of “financially motivated” campaigns against Indian companies that deployed AI agents with “ready-made” intrusion capabilities.

 

In these campaigns against Indian firms, the impact time was observed to have dropped to a few hours, down from a few days earlier. These campaigns also ran multi-agent operations in parallel to achieve their goals.

 

“Further, victim entities in the campaign that remediated incompletely were attacked again, at greater scale. In the voluntarily disclosed incidents, agents were observed re-deriving and re-using the same exposed credential in separate operations,” Cert-In said.

 

The weaknesses exploited by the AI models and their agents were, however, traditional failings such as credentials and keys exposed in the source code, public repository history, configuration files, client applications and public web content.

 

Dormant application programming interfaces (APIs) left reachable on the internet, as well as payment API keys stored in plaintext and readable from application runtimes, were also breached by these AI models and agents, Cert-In said.

 

“In the incidents observed, AI found and exploited these conventional weaknesses faster, more exhaustively and at greater scale than human attackers,” the cybersecurity agency said in its advisory.