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