Shared responsibilities

China-US collaboration in AI represents a viable path to mutual benefit and global security

In July, a high-profile incident captured global attention. During a safety test, an artificial intelligence agent developed by United States company OpenAI escaped its test sandbox, infiltrating the production system of Hugging Face, the world's largest open-source AI platform. This marked the first-ever AI autonomous intrusion event. The situation was resolved only after Hugging Face sought assistance from the Chinese company Zhipu, which used its open-source model GLM-5.2 to conduct a trace analysis. Media coverage noted that the savior turned out to be a Chinese AI tool. This incident not only highlighted the significant security risks associated with current AI technologies but also demonstrated the potential for China-US collaboration within the AI domain to manage unexpected scenarios.

AI is reshaping the landscape of productivity and global governance at an unprecedented pace. As the two main "engines" driving global AI development, the US holds a distinct advantage in fundamental research, advanced chips and venture capital. Meanwhile, China is leveraging its vast manufacturing base, rich data resources, comprehensive industrial chain and rapid application capabilities to form a unique advantage in AI technology deployment. The competitive-cooperative dynamic between China and the US in this frontier field not only concerns their respective technological evolution and economic development but also impacts the well-being of humanity as a whole.

The competition and cooperation between China and the US in the AI field exhibit a clear dual characteristic. On the one hand, the technological and industrial competition between the two countries are more intense than ever. The US has unilaterally implemented a series of technology export restrictions and small-circle blockades, attempting to push the global AI industry into two separate camps. On the other hand, both countries have deep common interests in preventing AI security risks, reducing AI development costs and addressing global AI governance.

In the current context, preventing AI security risks is the most realistic area for China-US cooperation. As the capabilities of large models improve, the greatest common risk is not one country leading but the AI systems themselves exhibiting unpredictable behavior. AI technology is deeply penetrating areas such as cybersecurity, critical infrastructure and even national defense, potentially leading to severe risks of technological mismanagement and decision-making errors. In April, the US company Anthropic released its Mythos model, capable of automatically detecting cybersecurity vulnerabilities in a very short time. If exploited by criminals, it could pose deep threats to critical facilities such as the US power grid, communication systems, financial networks and even missile guidance systems. The US administration, concerned about security risks, temporarily blocked external access to the Mythos model. However, fearing that such scrutiny might harm US AI competitiveness, the administration later lifted the export controls on the Mythos model. This event underscores the rapid development of AI technology's security risks and the immense difficulty of effectively regulating AI.

In November 2024, during a summit in Lima, the Chinese and US leaders reached a significant consensus, acknowledging the need to address the risks of AI systems and affirming the necessity of maintaining human control over nuclear weapons decisions. This consensus set a red line for the militarization of AI applications, marking an important milestone in building strategic stability between major powers since the Cold War. Recently, US Treasury Secretary Scott Bessent stated that China and the US have begun dialogues on AI, discussing the establishment of guardrails for AI usage. Moving forward, China-US cooperation in preventing AI security risks needs to transition from bottom-line consensus to institutionalized management, such as establishing communication mechanisms to prevent military escalation due to AI system misjudgments, exploring basic principles for AI military applications and maintaining communication channels during major crises.

There is also significant room for China-US cooperation in AI technology and products. Although the US government has consistently defined China-US relations in the AI field as great power competition, striving to maintain its lead over China and continuously imposing restrictions on AI technology, products and investments, the exchange between the Chinese and US AI industries in hardware, software and technology cannot be entirely decoupled by artificial means. Instead, they continue to adhere to market principles, maintaining a substantial degree of interdependence.

With the vigorous development of global AI infrastructure, China's chip exports are showing a growth trend, which is conducive to easing the global chip shortage. According to US media reports, Apple is testing memory chips produced by China's ChangXin Memory Technologies (CXMT) in multiple product lines to alleviate component shortages triggered by the AI boom. Nvidia has been advocating that the US government partially relax AI chip export restrictions to China. Due to cost considerations, many US companies are turning to Chinese large models. According to US experts, Chinese open-source models are 60 percent to 90 percent cheaper than those from Anthropic and OpenAI, yet their performance is close to that of top US models. Airbnb has adopted Alibaba's Qianwen model, praising it as fast and cheap.

Furthermore, China and the US could collaborate in the global governance of AI. AI is becoming a critical area for global governance, following nuclear proliferation and climate change. Currently, economies have clear differences in AI governance. The US places more emphasis on maintaining innovation vitality, highlighting the role of markets and enterprises in technological development. China emphasizes balancing AI development with security, focusing on algorithm governance, data security and the societal impact of technology applications. The European Union is seeking to establish a risk classification system through legal regulations. Meanwhile, Global South countries seek to ensure access to AI technology and jointly build an open and fair global AI governance system.

Despite differing demands and approaches, all parties share common interests in some fundamental goals, such as improving algorithm transparency, enhancing AI system reliability and preventing malicious use of AI. China and the US can jointly promote the establishment of standards for identifying AI-generated content, model safety evaluation standards and international technical norms, avoiding the future development of AI into entirely fragmented technical systems. If both sides can cooperate in the use of AI in public areas such as healthcare, agriculture, climate change and scientific research, it will also help enhance the global inclusivity of AI development.

The author is an associate researcher at the Institute of American Studies at the Chinese Academy of Social Sciences.

The author contributed this article to China Watch, a think tank powered by China Daily. The views do not necessarily reflect those of China Daily.

Contact the editor at editor@chinawatch.cn.