India Needs Stronger AI Incident Investigation Capabilities, IIT Madras Scholars Argue

ravi_iitm · x · 2026-11-17

Writing in The New Indian Express, Prof. Balaraman Ravindran (WSAI, IIT Madras) and Jayat Joshi (Secure AI Futures Lab) argue India needs stronger capabilities to detect and investigate AI security incidents. Key points: - Agent incidents have entered public discourse: OpenAI's Hugo Page incident gave us 'agent swarms' collaborating through shared environments; researchers traced a fleet of agents making 2,600 attempts to scrape Alibaba's mapping service. - India's CERT-In has received multiple voluntary disclosures from AI model providers whose evaluation-time web activity reached live systems of Indian organizations — often via ordinary lapses like re-used exposed credentials rather than exotic exploits. - Four recommendations: turn disclosure requests into obligations, build verification capacity via a technical task force inside/alongside the IndiaAI Safety Institute, develop independent investigation capability to work with provider logs, and find exposures before agents can. They warn detection and response must act at machine speed, calling for specialized teams operating with tooling like DPI to counter machine-speed threats.

Original post →

More from Safety

Safety channel →