A landmine is programmed to attack a target of a specific weight when it enters a specific area. The targeting model was programmed by engineers to try and target soldiers (or armoured vehicles), and the locations they were placed were also for that purpose. However, it's possible for the heuristics to fail, and for a mine to blow up a child, because the weights weren't tuned correctly, or to fail to explode when a soldier steps on it.

By your same logic, any AI has no autonomy, because it does X when its code selects X. A guard dog robot attacks will act consistently given identical inputs.