Breaking the Model:finalisation and a taxonomy of security attack

It is well known that security properties are not preserved by refinement, and that refinement can introduce new, covert, channels, such as timing channels. The finalisation step in refinement can be analysed to identify some of these channels, as unwanted finalisations that can break the assumptions of the formal model. We introduce a taxonomy of such unwanted finalisations, and give examples of attacks that exploit them.