Autonomous weapon systems (AWS) are not yet established, but strong military motives exist to develop and introduce them. A major fear is that they could deteriorate the stability in a crisis between two actors. To investigate this problem, we use agent-based modeling and simulation (ABMS). Here one AWS fleet patrols the perimeter of a protected area while a second moves outside. AWS of the first shoot if they sense being attacked or if AWS of the second enter the protected area. Errors in attack detection or in position determination can lead to escalation from crisis into war. The escalation probability is evaluated with a variation of these errors. Moreover, we elaborate on the shortcomings of ABMS in general.