As a core frontier of next-generation artificial intelligence, emergent intelligence in unmanned swarms achieves efficient collaboration and intelligent enhancement through biomimetic self-organization and local interactions, demonstrating significant potentials in both military and civilian domains. However, the reliability issues arising from emergence remain a critical bottleneck limiting its operational effectiveness. Guided by the practical need for unmanned swarm systems to evolve from functional realization to reliable emergence, this study analyzes their cross-layer coupling characteristics. Considering multiple constraints in complex dynamic environments, it outlines the emergent reliability issues and technical challenges across communication, perception, and behavioral layers. Based on a three-layer analytical framework of "information communication‒perceptual interaction‒swarm behavior", this study constructs a research system for the reliability of emergent intelligence. Furthermore, it progressively advances from exploring intrinsic mechanisms and evaluating emergence metrics to analyzing system reliability, proposing a new paradigm of reliability systems engineering for emergent intelligence, characterized by "theoretical modeling‒mechanism design‒evaluation and verification". This provides systematic theoretical, technical support and engineering solution references for the large-scale, reliable application of intelligent unmanned swarms in complex mission scenarios.