The classifier judged this item not relevant to AI security, so it is left out of lists, feeds, the API and dataset releases. If that is wrong, report a correction on this page.
InfoResearchPreprint
PIA: Agentic Physics-Informed Runtime Anomaly Detection
- Published
- Record updated
Summary
Runtime anomaly detectors for cyber-physical systems are either black-box models that miss stealthy manipulations or hand-built physical models that need expert effort per plant. The authors introduce physics-informed agents (PIA), where a large language model agent writes compact physical residuals and refines them against normal operating data. On the BATADAL water distribution benchmark, PIA with Claude Opus 5.5 reaches an F1 of 0.878 against 0.504 for the strongest of six baselines.