Mitigating covert compromises: A game-theoretic model of targeted and non-targeted covert attacks

Attackers of computing resources increasingly aim to keep security compromises hidden from defenders in order to extract more value over a longer period of time. These covert attacks come in multiple varieties, which can be categorized into two main types: targeted and non-targeted attacks. Targeted attacks include, for example, cyberespionage, while non-targeted attacks include botnet recruitment. We are concerned with the subclass of these attacks for which detection is too costly or technically infeasible given the capabilities of a typical organization. As a result, defenders have to mitigate potential damages under a regime of incomplete information. A primary mitigation strategy is to reset potentially compromised resources to a known safe state, for example, by reinstalling computer systems, and changing passwords or cryptographic private keys. In a game-theoretic framework, we study the economically optimal mitigation strategies in the presence of targeted and non-targeted covert attacks. Our work has practical implications for the definition of security policies, in particular, for password and key renewal schedules.