Information flow and probabilistic noninterference offer
popular notions of security properties that are sensitive
to the exact execution probabilities of the observable events,
though in practice it is virtually impossible to guarantee that
information flows only along authorized paths.
In this talk we relax the notion of probabilistic noninterference
through an approximate definition of process equivalence, by
means of which it is possible to measure the interference of
several classes of adversaries. For each class, which corresponds
to a specific security property, we also discuss the problem of
finding the most powerful adversary that maximizes the probability
that an external observer reveals an information flow.