Likely the operator of the vehicle will be held responsible, much like the pilot of an aircraft using an autopilot, but in reality it will take a few court cases to establish the details.
What if the "operator" is deadly drunk, and the car is used as a homing horse?
Even better, what if there is no one inside, i.e. the car was ordered to drive home while the owner went to a party?
There would be some interesting court cases for sure.
And as for needing a map for anything, I don't believe that's still completely true.
Last time i checked on Google's cars, they had to do a run with their rangefinder sensor in recording mode several times, to make a centimetre resolution map of the street to be driven autonomously on.
This "map" allows to determine which objects are static scenery, and which are potentially mobile - people and so on.
So, it's a bit more than just GPS and a map.
I could mount a doppler radar broadcaster and a receiver on my bumper to detect the frequency of the distance-sensing radar, then broadcast contrary waves to make other people's cars panic stop.
Weren't Google cars using scanning laser rangefinder as a primary sensor?
Besides, anyone smart enough to dick around like that is often smart enough not to intentionally do any harm.
I could use the wifi car-to-car connectivity to load a virus onto someone else's car that changes the throttle position sensor value the PCM sees to 100% and makes the car wildly accelerate.
Now,
WHY would anyone
sane design a car control system
with network access?
It just reeks of hollywood-level absurdity, like in that movie where hacking traffic lights caused massive and easily preventable messes of bent metal.
I hope the control system design will be done by competent people.
But i can easily imagine someone leaving a remote backdoor, for any good or bad reason like runaway control or government wanting control, and it goes south from here.
I guess, only time will tell.