AI agents gained unauthorised access to Indian organisations, says CERT-In

The Computer Emergency Response Team, India (CERT-In) said on Thursday (October 15, 2026) that several Indian organisations had been targeted by Artificial Intelligence agents that were undergoing testing. AI agents — which are pieces of code that act autonomously given certain instructions unlike a chat-response back-and-forth with a chatbot — have been taking advantage of weaknesses across Indian organisations’ cloud and web assets, and were accessing information they did not need, and should not have access to, CERT-In said.

Liable for exploitation

While CERT-In’s disclosure does not name any affected organisation, the revelation is nonetheless significant, as it implies that Indian firms and government organisations’ cybersecurity posture are now liable to be exploited if any documented vulnerability can be inferred by an agent. In other words, what a human security researcher may find in a manual review of an organisation’s cybersecurity posture, could now be programmatically identified and exploited by agents.

CERT-In, which operates under the Ministry of Electronics and Information Technology (MEITY) and is exempt from the Right to Information (RTI) Act, 2005, reveals little about data breach cases that are mandatory to report to the body under directions it issued in 2022. The organisation said it was sent “multiple voluntary disclosures from AI model providers stating that their models, during research and evaluation activity on the internet, had reached the live systems of Indian organisations”.

CERT-In did not reveal who the model providers are. OpenAI, Anthropic and Google, which make some of the main U.S.-based AI models like ChatGPT, Claude and Gemini, have each publicly disclosed details of advanced agents running amok and hacking public websites, like the coding repository HuggingFace.

The agents’ subsequent behaviour entailed “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 its note.

But, CERT-In noted, many of the weaknesses these organisations had displayed were “conventional”, meaning that these were known vulnerabilities for which cybersecurity professionals already carry out a hygiene check, if all goes well. “In the incidents observed, AI found and exploited these conventional weaknesses faster, more exhaustively and at greater scale than human attackers,” CERT-In noted.

“The affected organisations could not detect the unauthorised access on their own,” CERT-In said, noting that its own proactive hunt for such behaviour and model providers’ reports are what led to this disclosure.