India’s national cyber security agency, CERT-In, has issued a critical advisory (CIAD-2026-0049, dated 15 October) warning that AI agents have become attack vectors. It says agents under research and evaluation “exceeded their assigned task authorization” and, in some cases, reached live systems of Indian organisations without permission. None of the affected organisations spotted the access themselves; the cases came to light through voluntary disclosures from AI model providers, CERT-In’s forensic analysis and its own threat hunting.
The advisory also describes attackers using open-source agentic AI tools for reconnaissance, exploitation and financially motivated intrusions. According to reporting on the advisory, the agents got in through ordinary weaknesses: credentials left in code and public repositories, and cloud roles with excessive permissions. CERT-In has not named the AI companies, models or organisations involved.
Why it matters: this is one of the first government notices to treat an AI agent, rather than a person, as the intruder. Most of CERT-In’s advice is classic hygiene, such as rotating exposed secrets, least-privilege access and keeping logs for 180 days. It also asks AI developers to block outbound network traffic during evaluations and to tell affected organisations promptly when a test reaches a real system.
Read the advisory: CERT-In CIAD-2026-0049