Assessing risk from intelligent attacks: A perspective on approaches

Assessing the uncertainties in and severity of the consequences of intelligent attacks are fundamentally different from risk assessment for accidental events and other phenomena with inherently random failures. Intelligent attacks against a system involve adaptation on the part of the adversary. The probabilities of the initiating events depend on the risk management actions taken, and they may be more difficult to assess due to high degrees of epistemic uncertainty about the motivations and future actions of adversaries. Several fundamentally different frameworks have been proposed for assessing risk from intelligent attacks. These include basing risk assessment and management on game theoretic modelling of attacker actions, using a probabilistic risk analysis (PRA) approach based on eliciting probabilities of different initiating events from appropriate experts, assessing uncertainties beyond probabilities and expected values, and ignoring the probabilities of the attacks and choosing to protect highest valued targets. In this paper we discuss and compare the fundamental assumptions that underlie each of these approaches. We then suggest a new framework that makes the fundamental assumptions underlying the approaches clear to decision makers and presents them with a suite of results from conditional risk analysis methods. Each of the conditional methods presents the risk from a specified set of fundamental assumptions, allowing the decision maker to see the impacts of these assumptions on the risk management strategies considered and to weight the different conditional results with their assessments of the relative likelihood of the different sets of assumptions.