AEGIS: Adversarial Entropy-Guided Immune System
Thermodynamic State Space Models for Zero-Day Network Evasion Detection
As TLS 1.3 limits the effectiveness of traditional Deep Packet Inspection, content-reading classifiers like ET-BERT remain structurally vulnerable to byte-level adversarial morphing. We introduce a Thermodynamic Variance-Guided Hyperbolic Liquid State Space Model (TVD-HL-SSM) that discards payload bytes entirely in favor of 6-dimensional continuous-time flow physics. By projecting temporal features into a non-Euclidean Poincaré manifold and measuring sequence-wide Shannon entropy, AEGIS detects adversarial tunneling without ever reading what's inside the packet.
