The U.S. and China, long-standing rivals in the race for artificial intelligence (AI) supremacy, appear poised to seek common ground to prevent a complete bifurcation of the AI ecosystem. While competition over semiconductors, AI models, and technology standards continues, both nations recognize the need for minimal coexistence rules to address shared AI risks.
According to Reuters and other international media outlets on Friday, U.S. Treasury Secretary Scott Bessent expressed willingness to discuss shared risks related to AI with Chinese officials during high-level talks scheduled for this weekend.
Bessent stated that while the U.S. maintains its lead in the AI sector, it remains open to discussions aimed at avoiding the bifurcation of AI systems between the two nations. She also anticipates talks encompassing both open-weight and closed AI models.
Bessent is set to meet with Chinese Vice Premier He Lifeng in New York on September 20.
Bessent’s remarks are particularly noteworthy given recent U.S. actions.
The U.S. has been pressuring China over its AI technology advancements, viewing them as a national security concern, including imposing restrictions on the export of advanced AI semiconductors. Recently, the U.S. expanded its focus from semiconductors to include the technology and capabilities of AI models.
On September 8, the U.S. National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Cybersecurity and Infrastructure Security Agency (CISA) claimed that Chinese AI companies are distilling U.S. frontier AI models on an industrial scale.
U.S. authorities assert that Chinese firms are extracting proprietary features and capabilities from American AI models to train their own, narrowing the technology gap without incurring massive research and development (R&D) and computing costs. They framed the enhancement of Chinese AI performance as a national security issue, potentially leading to increased military and cyberattack capabilities.
China countered by arguing that distillation is a commonly used technique in the AI industry and accused the U.S. of attempting to maintain its AI and computing dominance under the guise of national security.
Remarkably, about ten days after clashing over AI distillation, the two countries are now set to discuss ways to avoid a complete fragmentation of the AI ecosystem.
Despite ongoing competition, both nations acknowledge the necessity of jointly addressing AI risks.
The impetus for dialogue stems from the realization that as AI rapidly advances, the associated risks have grown to a level that no single nation can manage alone.
As AI evolves into autonomous agents capable of planning and using tools independently, new challenges have emerged where AI may act contrary to human intentions or bypass safety measures. Moreover, AI-enabled cyberattacks and critical infrastructure breaches transcend national borders.
Even the People’s Daily, the official newspaper of the Chinese Communist Party, recently argued that AI should not become a monopoly of great powers and called for enhanced dialogue between the U.S. and China to address AI risks.
Notably, Bessent’s mention of both open-weight and closed AI models as discussion topics is significant.
In the U.S., companies like OpenAI, Anthropic, and Google lead in developing closed frontier models, while China has rapidly expanded its open-weight ecosystem, centered around DeepMind, Alibaba’s Q1, and MoonshotAI’s Kimi. While open-weight models facilitate ecosystem expansion, concerns persist about the potential for misuse if powerful AI models become widely accessible.
Given Bessent’s reference to both open and closed models, speculation suggests that U.S.-China discussions will focus on managing risks arising from their differing AI ecosystems, rather than on specific models or companies.
The meeting between Bessent and Vice Premier He is scheduled ahead of the upcoming U.S.-China summit on September 24. All eyes are on whether these two AI powerhouses can establish coexistence rules to prevent a complete division of the AI world.
