If the operators can select targets and the drone locks on and fires, that crosses into a high-risk moral gray-zone, since UIs are susceptible to misclicks and the drone may not consider collateral consequences (such as overpenetration.

If the drone can autonomously target and attack on its own algorithms, add to that the inevitable miscalculations meaning it will eventually kill a target that a soldier would not.

In the meantime anarchists amd revolutionaries should examine how to convice it it's been compromised to convince it to self-destruct.

And if it requires a signal from home to auto-destruct, how to block the affirmative signal.

I suspect the GLA is going to develop thick, sticky smoke bombs and signal jammers to make our ground drone blind and isolated. Then it can be neutralized and salvaged for parts.