Countries around the world are increasingly recognizing the importance of artificial intelligence technology and gradually becoming aware of the security challenges and ethical dilemmas it may pose. Encouragingly, various nations have achieved progress in the governance and risk management of AI technology, with numerous multilateral initiatives and mechanisms related to AI safety and security emerging rapidly.
Against the backdrop of ongoing developments in technology and governance, China and the United States launched the first governmental dialogue on AI in May. Following this dialogue, skepticism arose among some people about its outcomes, particularly because the U.S. continued to impose technological restrictions on China. Nevertheless, despite the risks associated with AI and the need to stabilize bilateral relations and promote global governance, dialogue between the two countries in this area holds great significance.
On one hand, China-U.S. cooperation will help better manage and address AI’s risks. Currently, both countries are intensifying their assessment of potential and real risks in the development and application of this technology. By attempting to craft timely responses from both technical and regulatory perspectives, both China and U.S. are accumulating significant practical experience. Additionally, given their differing approaches to AI research, application and regulation, their experiences are complementary.
From a technical standpoint, the U.S. holds advantages in foundational AI research and large model development, while China possesses richer usage cases and data accumulation. Collaboration between the two will facilitate the assessment and mitigation of inherent technological risks through model design and data training.
On the other hand, the dialogue is beneficial in stabilizing bilateral relations. Technology has become a core aspect of U.S. strategic competition, and if the two countries can communicate on AI issues to clarify their agreements and differences, it will help improve bilateral relations and ease tensions. While the dialogue may not resolve America’s “small yard, high fence” policy in the short term, the extensive list of AI applications makes discussions on security risks valuable in delineating risks and promoting new topics and in fostering mutual understanding through specific collaborative scenarios.
China-U.S. dialogue and cooperation on artificial intelligence also help convey positive signals to the rest of the world, mitigating new risks that may arise from great power competition over time. For instance, several European countries attach great importance to the technical rivalry between the major powers, believing that if the two cannot effectively manage technological competition, the overall efficacy of global AI governance will be weakened. Further, if the U.S. continues to form alliances that exclude China, the practice may result in incompatible technological development and lead to chaotic standards in governance, creating regulatory gaps.
The dialogue faces numerous challenges. Overall, the current atmosphere in China-U.S. relations is marked by heightened vigilance and a comprehensive suppression strategy by the U.S. in the tech sector, significantly dampening China’s willingness to achieve cooperative outcomes in AI. Although the U.S. hopes to separate the AI dialogue from its broader tech policies toward China, this contradictory approach — engaging in dialogue while simultaneously engaging in suppression — is unlikely to foster a positive environment for dialogue or enhance mutual trust.
Operationally, the dialogue also confronts three major issues: politicization of topics, obscuring of urgent risks and inconsistent expectations regarding dialogue outcomes.
First, unlike direct communication between technical personnel and scholars, intergovernmental dialogue often encounters the risk of politicization on AI issues. This prevents both sides from focusing effectively on the technology itself, shifting attention to the bilateral relationship beyond the technology.
Second, because of the increasing U.S. strategic competition against China, China-U.S. tensions have become an unavoidable topic in intergovernmental dialogue, making it difficult for both sides to advance discussions to address the common risks posed by AI. Instead, the dialogue tends to stall at bilateral restrictions and sanctions in the technological realm.
Finally, there are differences between China and the U.S. in assessing AI risks, international governance goals and domestic objectives, leading to varying expectations for dialogue outcomes.
In response to these challenges, the next phase of dialogue on AI should clarify goals, refine topics and fully explore the potential of multilateral and Track II dialogues. For governmental dialogues, both sides could authorize Track II dialogues or hold preliminary meetings at the governmental level before talks begin to clarify primary objectives and adjust mutual expectations.
Additionally, the next phase of dialogue could set a clearer and more defined agenda to eliminate the deviations caused by political factors and bilateral tensions, enriching the content of the dialogue while enhancing communication effectiveness. Both sides could incorporate more case studies on bilateral technical governance into the dialogue and include discussions on response strategies in specific scenarios, guiding the focus from bilateral disagreements to common challenges.
In setting topics, the two sides should consider dividing AI security issues into two categories: technological safety and application security. The former emphasizes risks that may exist at the technical level, such as algorithmic bias, model vulnerabilities and the endogenous risks posed by rapid model iterations. The latter focuses on the malicious and improper use of technology by non-state actors, social changes resulting from technological development and application, the militarization of artificial intelligence, incompatible governance systems and regulatory gaps.
On the technological safety front, China and the U.S. could first explore reliability, controllability and explainability of AI technology, discussing their respective concerns and expected solutions. Further, both sides could communicate on how to ensure non-discrimination in technology and safeguard humanity, addressing how to uphold scientific ethics and fundamental values during the development and training phases. Finally, they could discuss how to address potential technology control issues arising from the excessive development of generative AI and super artificial intelligence, exploring ways to balance technological advancement and effective regulation.
In application security, security risks can be broken into three categories: risks arising from the malicious use of technology by non-state actors, risks to social order caused by technological changes and security risks stemming from international interactions between states.
First, China and the U.S. could clarify the security threats posed by the malicious or abusive use of technology by non-state actors and explore how to establish “seatbelts” for the use of AI through values advocacy, legal regulations and technical constraints. Second, they could communicate about the potential risks that technological developments bring to social life, such as unemployment due to automation and the need for workforce retraining, aiming for better utilization of AI technology to empower economic development and facilitate industrial upgrading and labor transformation. Third, the two countries could assess the spillover risks from international interactions, exploring potential management and response strategies. Given the uncertainty surrounding the U.S. elections and potential adjustments in U.S. technology policy, it is essential for both sides to reach agreements on how to ensure timely intergovernmental dialogue on AI.
In addition to cooperating on risk management, the intergovernmental and Track II dialogues could also include discussions on how to harness the positive impacts of artificial intelligence. The two countries could collaborate on issues such as utilizing AI technology for global climate governance, controlling drug trafficking and other transnational crime and bridging the technology gap between countries.
On dialogue formats, they should emphasize multilateral dialogue opportunities and the potential of Track II dialogues, viewing China-U.S. AI cooperation from a multilateral security perspective. Track II dialogues should serve as an important complement to intergovernmental dialogues, with efforts made to involve more stakeholders.
China and the U.S. should adhere to multilateralism in AI governance, gradually establishing a governance mechanism with universal participation under the framework of the United Nations. As leading countries in AI technology, the two can work together within the UN framework to improve existing governance systems — which is significant for global governance of AI and conducive to creating a safe and orderly environment for future AI innovation.
Besides the UN mechanisms, the two should also leverage non-governmental multilateral dialogue platforms, such as scholar forums and technology summits. This approach helps avoid the disadvantage of focusing too narrowly on bilateral competition and promotes a broader and more comprehensive examination of the negative impacts of bilateral competition for both countries.
Additionally, China and U.S. should continue to recognize the potential of Track II dialogues, ensuring the stability of relevant mechanisms and further enriching the content by involving more individuals from the technology and business sectors. Track II dialogues should serve as a crucial complement to intergovernmental discussions, laying a solid foundation for government dialogue. For example, before initiating intergovernmental dialogue, China and U.S. could first engage at the Track II level to establish a preliminary consensus or clarify points of divergence. After the dialogues, Track II contacts could be used to review issues from the intergovernmental discussions, further enriching macro consensus and facilitating operational implementation.
Given that both China and U.S. have established governance institutions, mechanisms and research networks related to AI, they could also promote dialogue between these governance mechanisms and research networks in the future, creating more communication bridges for professionals engaged in technology development, risk prevention and applications. This would help both sides reach consensus on the risks associated with technology development and application, moving from dialogue to practical cooperation in the AI realm.
