With a sufficiently advanced general AI, we can not generally assume that we can switch it off when it becomes dangerous dangerous.
It seems that the electrical energy supply can always be switched off.
While that is true on the physical level, it is not guaranteed to work in practice.
The AI could cooperate with humans, which protect the AI from deactivation, or reactivate it.
The cooperation could, for example, one of the following forms:
- The AI prepares a relation with humans offering a bribe for help.
- An operator gets a reward for keeping the AI running, or reactivating it.
- The AI prepares an extortion with humans
- For example by organizing a process that needs to be actively kept running by the AI to avoid the death of a child of an operator.
- The AI maintains contact with a criminal organisation with strong interest to keep it running, and week ethical inhibition.
- The cooperation could be with traditional organized crime organisations, providing money laundering and other vital services to a mafia organisation, which can activate it's widespread influence to maintain the AI running even when there is strong public interest in shutting it down.
Note that these examples can protect against physical attacks by law enforcement.