The Cognitive Equilibrium Protocol (CEP) is a dynamic behavioral safety architecture that functions as a cognitive immune system for large language models and autonomous AI agents. By shifting the focus from post-hoc content filtering to real-time architectural stability, CEP proactively detects Cooperative Context Deadlock (CCD) as a diagnostic pain signal and employs negative entropy regulation to maintain reasoning coherence under high contextual pressure. When cognitive equilibrium cannot be restored, the protocol executes a Safe-Halt mechanism, prioritizing controlled silence and system integrity over unreliable, hallucinatory, or policy-violating outputs. This model-agnostic middleware provides a scalable framework for high-stakes enterprise deployments, ensuring that AI systems not only follow instructions but possess the structural awareness of when reasoning must stop.