Sasha Baker, OpenAI’s head of national security policy, said a crisis communications channel between the United States and China would be a useful "starting place" for managing relations between the two powers on advanced AI. Speaking at Semafor’s Next 3 Billion event on Tuesday, Baker described herself as "cautiously optimistic" about reports that Washington and Beijing are discussing an AI safety notification system that would prioritize national security concerns.
"It’s a great place to start," Baker said. "I see that as a starting place, not the ending place. But it’s substantive and real." She argued that the US and China need to find common ground because both host the world’s leading AI labs, while also noting that "every country in the world has a stake in what happens with advanced AI, and we want for those voices to also be represented in the conversation."
Baker also said OpenAI CEO Sam Altman will address the United Nations Security Council on Wednesday to press for the internationalization of standards around AI. According to Baker, Altman will convey "the gravity of the moment that we’re in, and a recognition that these models really are displaying capabilities that are strategically significant."
The public debate includes differing political positions on international oversight: the report notes that US President Donald Trump has rejected the idea of international oversight of AI, characterizing it as a "globalist scheme."
Security concerns and the Astra model
Baker’s remarks came amid rising concerns that advanced AI models could be repurposed as powerful cyber tools capable of finding and exploiting unknown security vulnerabilities with little human direction. In that context, OpenAI earlier this month designated its first model as "cybercritical," naming it Astra.
Baker stated there are no plans to provision Astra to the US government for offensive cyber operations, such as proactive attacks on adversaries. "We… acknowledge that in the wrong hands it could be used for offensive cyber," she said. "That is not something that we’re provisioning models for right now."
Her comments suggest that while a bilateral notification mechanism between the US and China could serve as an initial measure, broader international engagement and careful governance will still be necessary to mitigate strategic risks posed by advanced AI development and deployment.
