Afriat's Test for Detecting Malicious Agents

How can one detect if a set of agents is deliberately trying to avoid being detected? By assuming malicious agents are utility maximizers, we use a remarkable result developed by Afriat to construct a decision test that identifies such malicious agents with pre-specified Type-I error probability. Also a stochastic gradient algorithm is given to adapt the probe signals in real time to minimize the Type-II error probabilities of the decision test.