Actual Causes of Security Violations (Short Paper)

ABSTRACT. Accurate blame assignment for security violations is essential in a wide range of settings. For example, protocols for authentication and key exchange, electronic voting, auctions, and secure multiparty computation (in the semi-honest model) ensure desirable security properties if protocol parties follow their prescribed programs. However, if they deviate from their prescribed programs and a security property is violated, determining which agents should be blamed and appropriately punished is important to deter agents from committing future violations. This work proposes actual causation (i.e., identifying which agents' deviations caused a specific violation) as a useful building block for blame-assignment. The central contribution of this work is formalizing and reasoning about actual causation in decentralized multi-agent systems, in particular, to formalize programs (rather than events) as actual causes of security violations and to deal with non-deterministic systems.