Do you think AI governance should be technically rather than paper policy?