Checks and balances

Inclusion and openness are necessary foundations for global AI governance

Several key developments have suggested that artificial intelligence is quickly advancing. The solution to the Navier-Stokes Existence and Smoothness Problem, widely believed to be one of the most important open problems, is a hallmark development in AI for mathematics. Likewise, cybersecurity has seen rapid development, with Anthropic's Mythos showing the ability to detect and fix bugs in software. A few months later, an open-weight Chinese model called GLM 5.3 demonstrated a similar capability. Many believe that these advances are not just continuing, but accelerating, in a process described as "recursive self-improvement".

This progress in AI capabilities has sparked a debate over the structure of global AI governance. It has become increasingly clear that governments and institutions must help steer the trajectory of AI development. Some have advocated a unilateral approach to AI governance, in which some countries explicitly play a leadership role and other countries are relegated to a subordinate status. Other countries have advocated an egalitarian, multilateral system of AI governance. The increasing power of AI has made these debates increasingly urgent.

There are three key reasons why openness and inclusion are essential for AI governance: the importance of open-weight and accessible models, global openness in AI development and multilateral AI governance structures.

First, one key contention in AI has focused on whether AI models should be open or closed. Major AI labs of the United States, in particular OpenAI and Anthropic, have kept their models closed, meaning that the models can only be accessed through an interface called an API. At the same time, most of the top Chinese AI labs such as DeepSeek, Zhipu, and Moonshot have made their models open and free to be downloaded by all users. This allows the models to be run locally or by third-party providers.

In the area of cybersecurity, such openness has proved to be critical. This was demonstrated in an incident where an OpenAI model reportedly hacked into the servers of HuggingFace, an AI infrastructure company. When the company tried to use OpenAI and Anthropic's models to investigate and deal with the hacking attack, their requests were refused. This refusal is made possible by the closed nature of OpenAI and Anthropic's models, which allows these companies to decide when and how customers can use their models. Closed models often have restricted cybersecurity capabilities. However, HuggingFace was able to defend against the attack and investigate it by using GLM 5.3's cybersecurity capabilities.

Another benefit of open models comes from the standpoint of data sovereignty. If a model can only be accessed via an API hosted in a foreign country, countries are forced to give up their data in order to use that model's powerful AI capabilities, facing the risk of sudden service disconnection, as evident in the US government's intervention regarding Anthropic's Fable model. This threatens a country's sovereign control over its citizens' private and sensitive data, as well as important industrial secrets. While discussions will continue over the trade-offs between open and closed models in AI development, the openness of AI models has already provided substantial benefits to the global AI ecosystem. China's AI models have played a key role in keeping global AI models open.

Second, global openness and collaboration are of great importance to AI development. In the future, it is believed that the AI industry will be a key part of the global economy, perhaps its most important component. If the unilateral approach to global AI governance is implemented, this will present risks that some countries and segments of society will be left out of the growing economic prosperity generated by the development of AI. The inclusive approach to AI governance promises a future in which all countries and segments of society can share in the productivity gains and wealth generated by AI.

As an example of South-South AI collaboration, China's Huawei has bid to build Egypt's state AI data centers on its Ascend stack, and plans cluster-level Ascend 950 deployments in the Republic of Korea as a non-US option. China's joint development offers in AI are non-exclusionary, open and don't rule out forming partnerships and relations with other countries. At the World Robot Conference 2026 and the World Humanoid Robot Games 2026 in Beijing, China invited teams from 16 nations, including Japan, Brazil and Germany, to compete in fair and open contests. Such international collaborations and competitions let all countries learn from each other and advance robotic technologies for the benefit of all. By collaborating with countries, especially those in the Global South, not only do they maintain data and technology sovereignty and a full stack access to frontier models, they also gain bargaining power and capacity to make fair deals, which protect their national interests.

Third, AI governance structures should be multilateral, open and inclusive. An example is the World AI Cooperation Organization. WAICO's founding agreement states that membership is open to all nations without special entry hurdles. As a leading member of WAICO, China has made concrete commitments to support the organization and the other 28 member states. Among China's pledges tied to WAICO is the offer of 5,000 AI training and seminar opportunities for developing countries over the next five years. Another channel of global AI cooperation is the 2026 Singapore Consensus on Global AI Safety Research Priorities, drafted with input from leading researchers in China, the US, the United Kingdom, the European Union and Singapore, that maps shared priorities on agents, alignment, misuse and societal resilience. Safety risks from AI, even if they originate from a single country, can develop risks that concern the entire world. These include biosecurity and cybersecurity threats. For this reason, AI governance structures must take the lead in developing and coordinating global AI safety standards.

As the world's two leading AI powers, China and the US share a responsibility to build a constructive dialogue around key issues in AI governance. Through dialogue and consultation, they can highlight common interests and shared benefits from coordinating on safety and global AI diffusion, the importance of open models, global collaboration in AI development, and building multilateral AI governance structures. These initial conversations on open and inclusive AI development will play a central role in the future of AI development in the world.

The author is an assistant professor at the College of AI at Tsinghua University.

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.