Year
1990
Abstract
Employee screening and monitoring programs are an important component of the defense in depth at nuclear facilities. Ideally, such programs should maximize the probability of detecting an adversary and minimize the probability of falsely \"detecting\" a nonadversary by taking into account the characteristics of potential adversaries. This paper presents a model for identifying patterns of characteristics, or indicators, that may be more prevalent among adversaries than among non-adversaries and, therefore, can be used for detecting adversaries. It assumes that adversaries are detected when a background check or other monitoring program recognizes suspect patterns of indicators such as drug abuse, unexplained income, financial problems, abnormal scores on psychological tests, etc. The model quantifies the probabilities that adversaries will manifest various patterns of indicators. The analyses presented in the paper identify sets of patterns that are most likely to be manifested by several types of adversaries. Monitoring programs that can recognize these patterns should be more likely to detect adversaries. We have also estimated the probability that a normal employee might manifest the same patterns and be incorrectly detected as an adversary. The likelihoods use subjective probability judgements, however, statistical measures can be incorporated should they become available.