Database Activity Monitoring (DAM) systems are critical for detecting anomalous and malicious database inter- actions. However, traditional DAM architectures rely on direct inspection of SQL queries, exposing sensitive information such as schema details, user intent, and confidential data access patterns. This creates a paradox where security monitoring itself becomes a source of privacy leakage. In this paper, we propose a novel privacy-preserving DAM framework that eliminates exposure of raw SQL queries by transforming them into abstract metadata representations and deterministic query fingerprints. The system intercepts queries in a trusted execution environment, extracts structural features, and discards the original SQL before mon- itoring. We formalize privacy guarantees using an information- theoretic model and demonstrate that the proposed approach minimizes mutual information leakage between original queries and monitored data.