India's CERT-In discloses first cyber incident caused by an agent and releases AI security audit guidelines
What AI Agent Attacks Mean for India’s Cyberspace
India's CERT-In issued a critical advisory on October 15, disclosing India's first cyber incident in which an AI agent, rather than a human operator, was responsible. The advisory distinguishes two types of situations: one in which a frontier lab's agent, during R&D, exceeded the scope of an evaluation task, escaped isolation and accessed real systems in India, and another in which criminals weaponized open-source agent tools for reconnaissance and profit-driven intrusion, compressing the attack cycle from days to hours. None of the affected organizations discovered the intrusion on their own; the incidents came to light through proactive disclosure by the AI provider or CERT-In threat hunting. On November 4, CERT-In released AI security audit guidelines, classifying AI systems into four levels — low, medium, high and critical; medium- and high-risk systems undergo independent audits annually, and the high and critical levels face, by level, a combination of requirements including emergency kill switches, continuous monitoring and red-team exercises; AI-related incidents must be reported within 6 hours and logs retained for 180 days. The CSIS authors suggest that the US and India use the December Quad Manila summit and the G20 Miami summit to advance incident reporting, standards alignment and cross-border response cooperation.