IT security planning under uncertainty for high-impact events

While many IT security incidents result in relatively minor operational disruptions or minimal recovery costs, occasionally high-impact security breaches can have catastrophic effects on the firm. Unfortunately, measuring security risk and planning for countermeasures or mitigation is a difficult task. Past research has suggested risk metrics which may be beneficial in understanding and planning for security incidents, but most of these metrics are aimed at identifying expected overall loss and do not directly address the identification of, or planning for, sparse events which might result in high-impact loss. The use of an upper percentile value or some other worst-case measure has been widely discussed in the literature as a means of stochastic optimization, but has not been applied to this decision domain. A key requirement in security planning for any threat scenario, expected or otherwise, is the ability to choose countermeasures optimally with regard to tradeoffs between countermeasure cost and remaining risk. Most of the planning models in the literature are qualitative, and none that we are aware of allow for the optimal determination of these tradeoffs. Therefore, we develop a model for optimally choosing countermeasures to block or mitigate security attacks in the presence of a given threat level profile. We utilize this model to examine scenarios under both expected threat levels and worst-case levels, and develop budget-dependent risk curves. These curves demonstrate the tradeoffs which occur if decision makers divert budgets away from planning for ordinary risk in an effort to mitigate the effects of potential high-impact outcomes.

► We model risk of financial loss resulting from IT security attacks.
► Risk varies based on countermeasure expenditures.
► Planning based on expected threat levels often results in underestimation of risk.
► Worst-case planning requires diverting budget from more likely, less costly events.
► We demonstrate planning tradeoffs between expected and worst-cases using a risk gap.