It sort of turns the age old philosophical question on its head.<p>Instead of asking "Would you let the AI out of the box?" we need to start worrying about "Would someone even bother to put the AI in the box to begin with?".<p>The temptation from now on will always be to think "Well, the previous AI wasn't dangerous when we gave it access to the internet, so why should we be paranoid about this new one, which only has a couple of extra optimisations...".