The government’s A.I. Security Institute, staffed by alumni from OpenAI and Google, is becoming a model for countries grappling with A.I.’s emerging risks.