It doesn’t really specify anything. It is very vague, and interpretation is left to the regulator.
If the AI or whatever determines that you are impaired is offloaded to the cloud, then it can send a stop command remotely, and will almost certainly be abused by malicious actors (and some governments). If it is onboard, then it is still likely going to be a closed source software or hardware module that can send a stop command to the car.
In any case it creates a mechanism within cars for disabling it outside the control of the owner or operator
If the AI or whatever determines that you are impaired is offloaded to the cloud, then it can send a stop command remotely, and will almost certainly be abused by malicious actors (and some governments). If it is onboard, then it is still likely going to be a closed source software or hardware module that can send a stop command to the car.
In any case it creates a mechanism within cars for disabling it outside the control of the owner or operator