Unmanned aerial vehicles (UAVs) can be deployed to monitor very large areas without the need for network infrastructure. UAVs communicate with each other during flight and exchange information with each other. However, such communication poses security challenges due to its dynamic topology. To solve these challenges, the proposed method uses two phases to counter malicious UAV attacks. In the first phase, we applied a number of rules and principles to detect malicious UAVs. In this phase, we try to identify and remove malicious UAVs according to the behavior of UAVs in the network in order to prevent sending fake information to the investigating UAVs. In the second phase, a mobile agent based on a three-step negotiation process is used to eliminate malicious UAVs. In this way, we use mobile agents to inform our normal neighbor UAVs so that they do not listen to the data generated by the malicious UAVs. Therefore, the mobile agent of each UAV uses reliable neighbors through a three-step negotiation process so that they do not listen to the traffic generated by the malicious UAVs. The NS-3 simulator was used to demonstrate the efficiency of the SAUAV method. The proposed method is more efficient than CST-UAS, CS-AVN, HVCR, and BSUM-based methods in detection rate, false positive rate, false negative rate, packet delivery rate, and residual energy.