A compromised node in wireless sensor networks can be used to create false messages by generating them on their own or by falsifying legitimate messages received from other nodes. Because compromised nodes that create false messages can waste a considerable amount of network resources, we should detect them as early as possible. Existing studies for detecting such nodes can only be used in situations where sensor nodes do not move. However, it is possible that nodes move because of wind or other factors in real situations. We improve existing studies for detecting compromised nodes in mobile wireless sensor networks. In the proposed method, an agent exists on each node and it appends its ID and a k-bit code to an event message and the sink detects a compromised node by a statistical method. Our method can be used in static and dynamic environments. Simulations we conducted prove the effectiveness of our method.