Neuro-Agentic Control: A Deep Learning-based LLM-Powered Agentic AI Framework for Controlling Security Controls

Short summary

A neuro-agentic control framework couples an LLM-based planner (Gemini 2.5 Flash-Lite) with a Time-Series Foundation Model (TimesFM) to achieve physics-grounded autonomous defense of operational technology systems. A 'Counterfactual Physics Injection' mechanism simulates LLM-proposed interventions in the foundation model's latent space before actuation, rejecting hallucinatory or unsafe actions. On the SWaT industrial dataset, the framework prevented 33.3% of breaches versus 26.7% for LSTM and 13.3% for TCN, with zero physically invalid actions executed.

• •LLM planner + TimesFM foundation model for safe autonomous defense of industrial OT
• •Counterfactual Physics Injection rejects hallucinatory actions before actuation
• •33.3% breach prevention vs 26.7% LSTM and 13.3% TCN, with zero invalid actions

Generated with AI, which can make mistakes.