MAIN STORY
A necessary conversation with limited trust
At Hudson Institute’s Washington conference on AI power and national security, Caroline Zier, OpenAI’s deputy lead for national security, argued for engagement with China without treating dialogue as an automatic guarantee of cooperation. As reported by the South China Morning Post, she urged American officials to assess what information they could reasonably receive in exchange for what they disclose. Her position was not a demand to abandon diplomacy. Rather, she presented deliberate information-sharing as a prerequisite for diplomacy that can deliver practical results.
The first difficulty lies in the vocabulary of risk. For researchers, “safety” may refer to unpredictable model behavior, exploitable weaknesses and harm to users. For governments, the same expression can also carry questions about strategic advantage, surveillance and control over sensitive information. Sam Bresnick of Georgetown University expressed skepticism that an AI incident hotline would function effectively without prompt, candid communication. His observation is a warning about implementation, not proof that such a mechanism is destined to fail.
















