This paper considers what happens when a system erroneously places trust in an attacker. More precisely we consider untyped attackers inside a distributed system in which security is enforced by the type system. Our Key-Based Decentralised Label Model for distributed access control combines a weak form of information flow control with cryptographic type casts. We extend our model to allow inside attackers by using three sets of type rules. The first set is for honest principals. The second set is for attackers; these rules require that only communication channels can be used to communicate and express our correctness conditions. The third set of type rules are used to type processes that have become corrupted by the attackers. We show that the untyped attackers can leak their own data and disrupt the communication of any principals that place direct trust in an attacker, but no matter what the attackers try, they cannot obtain data that does not include at least one attacker in its access control policy.
展开▼
