Beijing should step in to fill the global “vacuum” in rules on artificial intelligence as a Chinese AI model opens up a policy divide in the United States, according to various institutions within China.
The calls came after Beijing-based start-up Moonshot AI launched its Kimi K3 model earlier this month, sending shock waves through Silicon Valley and the White House.
The model is reportedly almost as powerful as top US rivals and much cheaper.
Moonshot, which was founded just three years ago, released the model’s core components on the internet, allowing anyone to download, inspect and customise for free.
Kimi is an open-weight or open-source model, unlike many American closed artificial intelligence models, whose core inner workings are guarded closely by the company that developed them.
AI governance has become a new battleground for China-US rivalry and neither side has ruled out the possibility of dialogue on the issue.
The White House has reportedly been weighing a ban on US firms using Chinese open-weights, after Treasury Secretary Scott Bessent last week accused China’s AI companies of misappropriating American intellectual property and warned of potential sanctions.
That prospect of such a ban prompted pushback from a broad range of tech leaders and AI start-ups in the US.
However, in China, the Shanghai-based Zhangjiang Platform Economy Research Institute said that “in the long run, the real rivalry was about rule-setting power” and whoever had the best regulations first would have the edge.
“This round of events has exposed a number of unresolved blanks of rules in global AI governance, which have been precisely the areas occupied by political narratives,” the institute said in a social media post on Tuesday.
“The legal boundaries of model distillation, ownership of model outputs and standards for tracing AI supply chains’ origins – these rules are currently in a vacuum.
“Whoever is first to propose a solution accepted by most countries will hold the initiative in the next phase of AI governance.”
In an interview with Fox Business on Thursday, Bessent said the US government was finding the “watermarks” of large language models built in the US on many AI models coming out of China.
But the Chinese think tank argued that there was still no widely recognised technical method that could definitively prove that one model was trained on the outputs of another.
Another bone of contention in the US debate is model distillation, a technique that uses outputs from more powerful systems to train smaller ones. Bessent accused Chinese AI firms of conducting “covert, industrial-scale distillation attacks”.
Still, distillation is regarded as a common practice, particularly in the open-source community.
The Zhangjiang institute called for research on technology to trace the origins of models and the data used to train them, saying such efforts would support evidentiary needs while helping “gain an early advantage in setting international standards”.
It also called for continued support for Chinese firms and institutions investing in development of open-weight models, while shaping greater influence over issues such as open-source governance and the localisation of open-source licensing.
Yuyuan Tantian, a social media channel affiliated with Chinese state broadcaster CCTV, also said in a post on Monday that the choice between open-weight and closed models was no longer the most fundamental question.
Instead, it was “who gets to define AI’s safety, risks and future”.
“Once open-source and closed-source have both entered the AI industry, the question is no longer choosing one over the other, but how to use a single set of rules so that both models can play to their strengths within clear boundaries,” it said.
“This is also a way into understanding China’s governance position.
“China supports openness, but that doesn’t mean it advocates unconditional proliferation of all capabilities.”
Citing people who have long worked on AI policy research, Yuyuan Tantian said China could adopt a governance approach of opening up foundational capabilities, limiting access to some frontier capabilities, keeping commercial services closed and imposing access controls and safety evaluations for high-risk capabilities.
It also called for clear lines around alleged “infringement”, noting countries should establish mechanisms for reporting major security incidents and handling vulnerabilities together within an open AI ecosystem.
“Whoever can make AI more inclusive and high-risk capabilities more controllable will be better positioned to define AI’s next phase,” it said.
At the opening of the 2026 World Artificial Intelligence Conference in Shanghai on July 17, Chinese President Xi Jinping pitched China as a committed provider of international public goods in AI, a field where the country is a front runner, while vowing to build a “just and equitable” system of global governance on the technology. -- SOUTH CHINA MORNING POST
