OpenAI outlines technical controls and veto guidelines for frontier AI safety