As AI systems become increasingly embedded in multi-tenant platforms, ensuring regulatory compliance and accountability requires more than ad hoc rule encoding. In the absence of a formal mathematical foundation, compliance logic cannot be guaranteed to remain consistent under evolving legal norms. This paper introduces a decision-theoretic semantic compliance framework that integrates a rigorous probabilistic reasoning model with ontology- and rule-based representations. Legal obligations are encoded in OWL ontologies and SWRL rules, while compliance judgments are derived through a mathematically grounded chain that includes prior probability estimation, Bayesian updating, likelihood ratio testing, and log-likelihood ratio decomposition. The architecture is modular and extensible, encapsulating compliance logic in a semantic layer that can be updated independently of the core platform. A prototype implementation in a healthcare scenario demonstrates the framework’s ability to detect policy violations, provide interpretable reasoning traces, and adapt to jurisdiction-specific regulations. The approach offers a verifiable and reusable compliance model applicable to a wide range of high-stakes, regulation-intensive domains, enhancing both transparency and trust in AI-enabled decision-making.
Jingyuan Xu (Tue,) studied this question.