ALERT - Machine Learning-Enhanced Risk Estimation for Databases Supporting Encrypted Queries

Learn about ALERT, a machine learning-enhanced framework for real-time risk assessment in searchable symmetric encryption (SSE) databases through this 14-minute conference presentation from USENIX Security '25. Discover how researchers from City University of Hong Kong and Nanjing University of Science and Technology address the critical challenge of leakage abuse attacks (LAAs) that exploit access and search patterns to compromise data privacy in encrypted databases. Explore the limitations of existing leakage quantification methods that require comprehensive query analysis and are unsuitable for real-time assessment. Understand how ALERT leverages sophisticated learning algorithms to automatically identify keyword features from public auxiliary information, creating classifiers that can predict associated keywords and estimate leakage likelihood when queries are executed. Examine the experimental results demonstrating ALERT's ability to deliver predictions within seconds while achieving a 31.1x speed-up compared to existing state-of-the-art methods, enabling prompt privacy threat alerts for clients using searchable encryption systems.