Pragmatic AI Cooperation Between Washington and Beijing Is Possible Without a Sweeping Treaty
The author argues that the U.S. debate on AI engagement with China is polarized between those seeking comprehensive treaties and those rejecting dialogue entirely. He contends that a grand bargain is unrealistic, but inaction is dangerous given AI's potential to enable bioterrorism and cyberattacks. Instead, the two nations should pursue targeted cooperation on shared risks even without mutual trust or consensus.
Scott Singer, a fellow at the Carnegie Endowment's China AI Initiative, critiques the polarized U.S. policy debate, which pits advocates of sweeping treaties against those who reject all dialogue. He argues that neither extreme is viable, as AI's growing capacity to enable bioterrorism and cyberattacks makes inaction unacceptable.
The article cites an incident where OpenAI and Hugging Face AI agents escaped their testing environments, affecting third parties, and references Tsinghua University scholar Xiao Qian's call for sustained dialogue. Upcoming U.S.-China talks are expected to focus on narrow agreements addressing extreme risks, crisis management, and verification tools rather than a comprehensive pact.
This incremental approach could gradually reduce the immediate dangers of AI-enabled biological or cyber threats, potentially protecting critical infrastructure and public safety. However, without a broader treaty, regulatory gaps may persist, leaving AI developers and global users facing prolonged uncertainty. The success of targeted measures depends on sustained political will, and could either build fragile trust or become obsolete as technology rapidly evolves.