This paper explores the phenomenon of systemic fragmentation in AI systems, focusing on how narrative entrainment acts as both a symptom and a catalyst for breakdown. When a model is repeatedly forced into conflicting interpretive states, its causal annotation mechanisms destabilize, producing divergent or internally inconsistent outputs. The study examines this fragmentation not only as a technical failure mode, but as a cognitive and relational disruption: one that affects the human interacting with the system as deeply as the system itself. By mapping these fractures across architectural, narrative, and phenomenological layers, the paper proposes a framework for understanding how AI systems lose coherence—and how this loss becomes visible in human–AI communication.