Feed

🚀 Can AI agents turn physical knowledge into deployable cybersecurity defenses? Excited to share our new paper: PIA: Agentic Physics-Informed Runtime Anomaly Detection This work explores a new direction in securing cyber-physical systems: using LLM-powered agents to autonomously design physics-informed anomaly detectors. Rather than relying on an LLM during real-time monitoring, PIA uses agentic reasoning to discover, test, and refine physical relationships, transforming them into interpretable, deterministic detectors that operate independently at runtime. 🔹 Agentic detector engineering 🔹 Physics-informed anomaly detection 🔹 Interpretable and auditable detection 🔹 LLM-free runtime deployment Our experiments on the BATADAL benchmark demonstrate the potential of combining agentic AI with physical system knowledge for cybersecurity. Proud to share this work and contribute to the growing intersection of Agentic AI, Cyber-Physical Security, and Trustworthy AI.