Enterprise knowledge systems confront an inherent conflict between robust contextual comprehension and rigorous security constraints. This research presents a theoretically grounded security-preserving context-aware retrieval architecture that fundamentally reimagines how Retrieval-Augmented Generation (RAG) can operate within strictly controlled information environments. We present two principal innovations: (1) a hierarchical security-aware embedding space that mathematically preserves access control constraints during semantic search, and (2) an adaptive re-ranking algorithm that optimizes the relevance-security tradeoff through a novel information utility function. Our theoretical contributions are validated through a reference implementation deployed across three distinct enterprise environments with varying security requirements. Experimental results demonstrate statistically significant improvements in both precision-recall metrics and security compliance compared to traditional approaches. We provide formal proofs of security guarantees under specific threat models and release anonymized benchmark datasets. This research advances the fundamental understanding of how contextual reasoning can be preserved while maintaining strict information boundaries, contributing broadly applicable principles to the fields of secure information retrieval and enterprise AI systems.