US and China Face Off on Global AI Safety Rules
As Washington and Beijing prepare to meet, the path to global AI safety hinges on a fragile cooperation between rivals who view each other as primary threats.
Efforts to establish global standards for artificial intelligence safety are increasingly dependent on cooperation between the United States and China. However, this dynamic is complicated by deep mutual suspicion, with both superpowers viewing the other’s AI strategies as a direct national security threat. As tensions rise over the rapid advancement of the technology, the prospect of a unified global approach remains tenuous.
AI governance is expected to be a central topic in upcoming talks between President Donald Trump and Chinese leader Xi Jinping in Washington. American tech leaders argue that preventing worst-case scenarios requires some level of engagement with Beijing. Yet, political realities suggest that any resulting agreements may be superficial, serving more as a starting point for future dialogue than a binding framework for safety.
Competing Strategies for AI Dominance
The two nations are pursuing AI leadership through fundamentally different methods. U.S. companies focus on high-performance computing power and proprietary, closed-source models, supported by rapidly expanding data center infrastructure. In contrast, China is emphasizing the global adoption of its open-source, more affordable AI models. This divergence creates friction, as U.S. restrictions aim to keep China from accessing the world's most advanced chips, while Chinese models gain traction even within American firms.
U.S. Treasury Secretary Scott Bessent has stated that America needs to develop more open-source models to counter China’s influence. Meanwhile, the U.S. government has accused China of attempting to extract capabilities from American systems, while Beijing accuses Washington of trying to suppress Chinese competitors to build a monopoly. These opposing narratives make technical cooperation difficult, even as both sides recognize the existential risks posed by unregulated AI.
Verification Challenges Hinder Trust
A major obstacle to agreement is the inability to verify compliance. David Sacks, a top AI adviser to the White House, has argued that unlike nuclear treaties, AI commitments cannot be effectively monitored or verified. He warned that trusting China to slow down development is impractical because there is no mechanism to confirm if they are honoring any promises. This lack of verifiability undermines the foundation of any potential treaty or safety accord.
Industry Calls for Cautious Cooperation
Leading AI firms like Anthropic and OpenAI find themselves in a complex position. While they assert that the U.S. must maintain its technological lead for national security reasons, they also advocate for cooperation with China to manage safety risks. Anthropic CEO Dario Amodei recently called for a slowdown in AI development to ensure safety, suggesting that the U.S. should continue to restrict China’s progress while engaging in dialogue to prevent catastrophic outcomes.
According to reporting by GN technics/ai (en-US), experts suggest that a formal consensus may be impossible, but political space can still be created. By acknowledging that AI poses risks to both countries, leaders can establish a baseline for dialogue. This approach prioritizes managing the balance of power and safety over achieving a perfect regulatory agreement, recognizing that the stakes of failure are too high for either side to ignore.
