Safe abstractions of data encodings in formal security protocol models

When using formal methods, security protocols are usually modeled at a high level of abstraction. In particular, data encoding and decoding transformations are often abstracted away. However, if no assumptions at all are made on the behavior of such transformations, they could trivially lead to security faults, for example leaking secrets or breaking freshness by collapsing nonces into constants. In order to address this issue, this paper formally states sufficient conditions, checkable on sequential code, such that if an abstract protocol model is secure under a Dolev-Yao adversary, then a refined model, which takes into account a wide class of possible implementations of the encoding/decoding operations, is implied to be secure too under the same adversary model. The paper also indicates possible exploitations of this result in the context of methods based on formal model extraction from implementation code and of methods based on automated code generation from formally verified models.