There is no simple way to shut down advanced AI once it exists, so societies must learn to manage its risks rather than rely on a single fail-safe.