ECA–Cybersecurity: A Unified Endogenous Architecture for the Cybersecurity, Resilience, and Governability of Human–AI Systems

This scientific corpus presents ECA–Cybersecurity, a unified theoretical, structural, methodological, and operational research framework dedicated to the cybersecurity, resilience, and governability of complex adaptive artificial intelligence systems and coupled Human–AI environments. The architecture integrates Crowd-Based Dynamics (CBD), Endogenously-Constrained Artificial Intelligence (ECA), UCQ, DUAL, RAG-RES, and RES–RAG, establishing a common analytical framework for examining the propagation of disturbances, structural memory, accumulation, saturation, conditional and activated divergence, reversibility, and the progressive loss of governability. Its central scientific proposition is that cybersecurity cannot be reduced to the prevention of unauthorized access, the detection of isolated attacks, or the protection of technical infrastructures. Security must also encompass the preservation of the structural conditions that allow an artificial system to remain observable, traceable, correctable, interruptible, recoverable, and subject to legitimate human control. The corpus distinguishes three interdependent dimensions of cybersecurity: technical security, semantic security, and structural security. It examines how adversarial manipulation, prompt injection, retrieval poisoning, persistent memory contamination, unauthorized tool execution, and multi-agent propagation may alter the internal dynamics of artificial systems, even when their observable functioning appears coherent. The CBD framework provides a structural interpretation of these processes through accumulation, propagation, memory, saturation, volatility, temporal dynamics, and regime transitions. ECA introduces endogenous security constraints designed to preserve operational boundaries, functional integrity, safe alternatives, recovery mechanisms, and human intervention capacity. The RES–RAG paradigm contributes a complementary semantic dimension, examining the relationships between human intention, contextual reconstruction, semantic divergence, and generated responses. The corpus further studies semantic attractors, semantic filters, the Semantic Citadel, and the Coherent State Network Protocol (CSNP) as candidate mechanisms for protecting semantic coherence, critical invariants, and authorization boundaries. UCQ and DUAL contribute to the structural interpretation of states, latent conditions, observable manifestations, and dynamic transitions, while RAG-RES addresses knowledge reconstruction, informational continuity, and traceability. The research also develops a structural approach to security governability, incorporating system observability, operational traceability, accessible safe trajectories, recovery capacity, endogenous constraints, and residual human control. Particular attention is given to structural thresholds, conditional reversibility, temporal persistence, memory effects, regime changes, and the distinction between local disturbances and systemic transformations.