||While the connected car is predicted to prevail by 2025, the appearance of novel cyber attacks is concerned. In this paper, we especially focus on the false data injection attack to such a service as collects information from the cars, analyzes the information, and drives the traffic society based on the result. The attack is such as follows: given a service which, for dynamic route selection, collects travel time from cars, calculates the data, and distributes the average travel time for each road, (1) malicious cars send false data to the service, (2) the service system calculates the wrong average travel time, (3) the system distributes the result, (4) the cars which believe the information select wrong routes, (5) finally, a traffic jam occurs. To examine the countermeasure technology of such unknown attack, the generation and analysis of the attack data are necessary. Furthermore, in order to reduce arbitrariness and cost in generating the data, the automatic generation is preferable to the manual one. Thus, in this paper, by using reinforcement learning, we developed a technique which automatically generates attack data inducing congestion on a vacant road.