Explainable Privacy-Budget Governance for Multi-Service Educational AI Systems

Abstract

The rapid expansion of AI driven educational platforms, such as learning management systems, automated assessment tools, and learning analytics services, has intensified concerns surrounding privacy, fairness, and institutional governance. Although differential privacy (DP) is increasingly used to safeguard student data, current educational AI systems lack an explainable, governance-aware allocation framework for allocating privacy budgets across heterogeneous services operating under a shared privacy constraint. This paper introduces a governance‑oriented reformulation of existing differential privacy budget‑allocation approaches for multi‑service educational AI systems. Rather than proposing a new differential privacy mechanism, the approach embeds institutional policy constraints, fairness considerations, and interpretable decision rules directly into the privacy‑budget allocation process. Each educational AI service is assigned a local privacy budget within a constrained global privacy limit (ε̄), determined according to its pedagogical relevance, sensitivity to fairness disparities across student groups, and governance‑defined explainability requirements.

The allocation process is formulated as a constrained multi‑objective decision framework, where trade‑offs among pedagogical utility, fairness sensitivity, and governance transparency are resolved through rule‑based policies rather than black‑box optimization strategies. An explainability layer produces auditable allocation justifications, counterfactual policy analyses, and governance logs designed to support institutional oversight. A structured comparative analysis evaluates the proposed approach against uniform, utility-only, and non-explainable allocation strategies using governance-relevant criteria such as policy traceability, fairness awareness, and auditability. The results demonstrate that the proposed framework provides stronger governance alignment and transparency, thereby bridging the gap between high-level AI governance principles and operational privacy management in educational AI systems