Killbots disrupt the suspension of disbelief that we currently use to excuse many killings by law enforcement: We find it morally acceptable for someone to use deadly force to protect themselves, and we use this to shortcut a more extensive analysis of the morality of killings by law enforcement.
In many cases the deadly force wasn't actually necessary or the need was so absurdly induced by the wielder that they should be culpable by the outcome. But the line is really hard to see when you weren't there, in their shoes, so we tend to defer judgement. At the very minimum this does provide a limited check on the use of deadly force.
For a killbot it couldn't be more clear: the operators life was not at risk, erasing the fiction that it was automatically excusable for personal protection and forcing us to justify the killing on a more complex basis.
Not that we can't-- I think sometimes we could, but probably in many situations we can't. E.g. right now, a gunman "safely" barricaded in a house eventually gets stormed in and killed. In a killbot equipped world, I don't think we could morally send in a killbot-- I think we'd have to wait for them to try to escape at least. And if we couldn't kill them with a killbot, well then that opens questions about the morality of doing the same thing with people.
I'd feel better about bots with less lethal weapons: nets, incapacitating gasses, tranquilizer darts, stun guns, etc. They may be radically less effective but one of the advantages of keeping the operator safe from harm is that a less effective weapon is easier to justify: its failure won't get the operator killed.
Aside, -- the specific application here with an interchangeable cartage just sounds like asking for trouble when someone leaves the wrong munition loaded.