One application forseen (And already indeed in use) for robots is security. Security can be a fairly high risk job, and with some less savoury or more determined characters- you may be putting yourself directly in the firing line.
Now the thing is - you cannot expect security robots to be too clever, and at their safest - little more than mouthy equivalents of security dogs. While I am a robot labour advocate - I would not want robots touting guns. Although I am not really a great believer in things ending up like The Cylons or The Terminator - all it would take is a simple bug (not AI but simple programming mishap) for the gun to fire when not intended (I don’t mean monster - I mean simply malfunction). Something like that would need extensive testing before being deployed - and also a method for a security control room (which would still be required) to be alerted before such events (ie a closed circuit that only allows a weapon deployment after such notification - and that allows override from the control room).
Saying that - there are systems - like the Rotundus. Some security robots are simply for surveillance - cameras on mobile platforms. If the camera is able to recognize an intruder, then it can alert the authorities and take a nice snapshot, or point the current security monitor feed at what they have caught. They can also - as said before - be very mouthy, and try to deter the intruder - inform them that the authorities have been called, that they are now being watched, that there movements have been recorded and are now stored at a secure location. The robot may even be able to try and discretely get some form of biometric identification so the intruder can be positively ID’d.
One problem is of course the robot may be then attacked- but this is much better than a human being attacked, and besides- it the surveillance data has already gone off to a remote location. If the robot is however more expensive than the equipment being protected - then it is a matter of serious consideration.