So far, we modeled channels relationally, taking into account that the same channel inputs may produce possibly different outputs. In this chapter, we refine the channel model to take into account the probabilities of the channel outputs. This is how channels are modeled in information theory and used in statistical inference. Bayesian reasoning allows for inverting channels and calculating the probabilities of channel inputs corresponding to a given channel output. Such inversions bring into focus the problem of secrecy and confidentiality, since the attacks on such properties are channel inversion tasks. The discussion about secrecy opens up the vast area of cryptology.