Probabilistic Opacity in Refinement-Based Modeling

Given a probabilistic transition system (PTS) partially observed by an attacker, and an -regular predicate over the traces of , measuring the disclosure of the secret in means computing the probability that an attacker who observes a run of can ascertain that its trace belongs to . In the context of refinement, we consider specifications given as Interval-valued Discrete Time Markov Chains (IDTMCs), which are underspecified Markov chains where probabilities on edges are only required to belong to intervals. Scheduling an IDTMC produces a concrete implementation as a PTS and we define the worst case disclosure of secret in as the maximal disclosure of over all PTSs thus produced. We compute this value for a subclass of IDTMCs and we prove that refinement can only improve the opacity of implementations.
View on arXiv