Long Short-Term Memory and Monte Carlo Tree Search for Proactive Intrusion Detection and Predictive Defense

This paper presents a multi-layered agentic AI framework for proactive cybersecurity defence, integrating an LSTM Autoencoder for zero-day behavioral anomaly detection with a Monte Carlo Tree Search (MCTS) module for predictive adversary forecasting. The system follows a modular architecture aligned with the OODA loop, enabling autonomous threat response. Benchmark results demonstrate 98.5% detection accuracy and a mean time-to-response under 60 seconds, a significant improvement over the manual baseline of 15–20 minutes.