Answers are grounded in this paper and cite their sources.
The Biba integrity model is a foundational access-control mechanism for preserving data integrity in enterprise security architectures. However, although its limited confidentiality protection is well recognised, its exposure to inference-driven threats remains insufficiently formalised. In Biba-based environments, adversaries may exploit semantically related, structurally overlapping, and sequentially composed authorised interactions to infer sensitive information without directly violating access-control rules. This paper formalises inference risk as cumulative information leakage across authorised observations using L n = ∑ i = 1 n I ( Y ; X i ∣ X 1 , ⋯ , X i − 1 ) , where Y denotes a sensitive target and X i denotes the i th authorised observation. This formulation shifts security analysis from isolated access-policy compliance to history-aware evaluation of how sensitive knowledge accumulates over time. Building on this model, the paper proposes the Inference-Aware Data Protection System (IADPS) , which integrates lattice-aware restrictions, Secure Query Validation Service (SQVS) logging, CRF-based sequence modelling, Histogram Gradient Boosting, CRF-probability augmentation, cumulative leakage monitoring, adaptive differential privacy, enterprise-rule monitoring, conservative fallback, and auditable response control. Evaluation uses an actual synthetic HR3000 dataset and an augmented benchmark HR dataset. The synthetic results demonstrate strong learnability under controlled conditions, with several non-linear, neural, and hybrid models achieving perfect mean performance. On the more heterogeneous benchmark dataset, the proposed CRF–HGB–CRF-Probability model achieves the strongest overall multiclass performance, with a mean accuracy of 0.9034, F1-score of 0.7033, and ROC–AUC of 0.9841. Enterprise-monitoring results further show that no single mechanism dominates sensitivity, precision, ranking quality, and alert volume: the hybrid provides the highest recall, the rule-based monitor the highest F1-score, and the confidence-based monitor the highest detection ROC–AUC. The leakage analyses show that authorised interactions progressively increase inference exposure. IADPS reduces cumulative leakage under selected model-specific configurations, decreases confidence and entropy leakage by approximately 65%, and reduces probability-derived leakage while preserving the primary predicted-class output. Robustness experiments show gradual degradation under missing data, feature noise, and distribution shift, while SHAP analysis identifies privilege overlap, privilege diversity, entropy, CRF probability features, temporal privilege behaviour, and uncertainty measures as principal explanatory factors. Overall, IADPS provides a theoretically grounded and operationally motivated approach to inference resilience by combining structural restrictions, hybrid modelling, cumulative leakage accounting, adaptive privacy control, enterprise-monitoring integration, explainability, and conservative fallback. The results support its feasibility as a layered software-based monitoring framework, while also showing that production deployment requires explicit inference-risk labels, calibrated risk scores, model-specific mitigation, threshold governance, and human oversight.
No takes yet. Share an insight, caveat, or question.
Ayamga et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: