The report examines to what extent a tactical decision support system can be used for planning underwater surveillance without making the surveillance predictable. The issue has been explored using a computer prototype that generates search patterns for active sonars, with the help of machine learning techniques derived from poker bots. The unpredictability of the prototype has been studied both theoretically and empirically. The results of the study indicate that an adversary, that studies surveillance activity over time, cannot predict and exploit the search patterns suggested by the prototype. The conclusion from the study is that decision support based on AI for pokerlike games is able to generate unpredictable search patterns for active sonars.