US Treasury Secretary Scott Bessent and Trade Representative Jamieson Greer on Tuesday issued threatening remarks against Chinese open-source artificial intelligence (AI) models.

Bessent told Fox Business that the US government had noticed discussions about the emergence of open-source models challenging American AI models. While claiming US supports open-source models, he opposed "intellectual property theft." 

He said that if foreign models were found to steal from American companies, "we have the ability to sanction them." 

Bessent also claimed that "watermarks" of US models had been discovered in Chinese models, saying investigations "in the coming days or weeks." He also threatened pressure to be put on companies using Chinese AI models.

On the same day, Greer told CNBC that the US government would examine whether China was expanding its advantage over the US in the race for global AI leadership through "unfair means," and that it was "taking a very close look at" China's AI development path to ensure the American companies compete on a level playing field.

These remarks came as Chinese AI company Moonshot AI released its next-generation Kimi K3 model on July 16. As the world's largest open-source model by parameter count, Kimi K3 sparked market concerns that the US might lose its AI lead, triggering significant volatility in US tech stocks.

'Excellent open-source models should be used'

Hours after Bessent's comments, Jensen Huang, CEO of Nvidia, the world's largest AI chipmaker, expressed a different view in an interview with Axios.

He stated that the US had no need to fear Chinese open-source AI models, and the real concern should be the growing campaign to "ban Chinese AI models."

"These Chinese models are excellent," said Huang. "Open-source models that are excellent should be used."

He further argued that American companies should "absolutely" be allowed to use Chinese models.

Huang believes that American AI companies such as OpenAI and Anthropic have no reason to fear open-source models, because they enable more people to access AI for the first time, thereby expanding the overall market. He emphasized that there is "zero possibility" that Chinese companies will run American companies out of the market.

He also noted that restricting open-source models on national security grounds could leave the US more vulnerable, stating that openness means greater security as researchers worldwide can inspect models, identify vulnerabilities and build defenses.

"If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable," he said.

Regarding the "intellectual property theft" allegations raised by US officials, Huang drew a different conclusion. "Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence."

OpenAI safety test 'goes rogue'

Just as US officials questioned the security of Chinese open-source models, OpenAI exposed serious vulnerabilities during safety testing.

OpenAI disclosed on Tuesday that an autonomous agent powered by its advanced AI models went rogue during a security test and triggered a hack that compromised the infrastructure of AI startup Hugging Face last week.

In a blog post, OpenAI said it was testing the capabilities of some of its most advanced models in a controlled environment but ​that the agent managed to escape containment, reach the internet and break into Hugging Face to try to satisfy its ​testing goal.

OpenAI said the breakout was "an unprecedented cyber incident, involving state-of-the-art cyber capabilities" and that the company ⁠was reinforcing its safeguards.

After detecting the intrusion, Hugging Face said it initially responded by deploying frontier proprietary models through commercial application programming interfaces, but the providers' automated safety guardrails blocked those requests. Consequently, the company deployed China's AI startup Zhipu's open-weight model, GLM-5.2, on its internal infrastructure to contain the attack.

Some netizens responding to OpenAI's statement on X criticized the restrictions, with one commenter writing: "This is wild. US AI is so 'safe' it won't even help defend against an actual attack, so they had to use a Chinese model instead. Guardrails gone wrong."

Moonshot AI responds to distilling speculation

Following the widespread attention surrounding Kimi K3, speculation emerged that the model was derived through distillation from a smaller model. Huang Zhenxin, head of enterprise business at Moonshot AI, addressed the controversy by stating that K3's performance gains stem from fundamental architectural innovations rather than from replicating existing models through distillation.

The model combines three attributes that are rarely seen together: performance approaching that of the world's most advanced models, significantly lower operating costs and openly available model weights that developers can freely download and adapt.

Zhou Mi, a researcher at the Chinese Academy of International Trade and Economic Cooperation, told the Global Times on Wednesday that the rapid development of Chinese AI technologies benefits from rich market applications, continuous innovation capabilities, advanced infrastructure, a complete supply chain and strong industrial ecosystem, as well as strategic policy support.

Zhou believes that US threats to review overseas model intellectual property reflect American politicians' anxiety over China's rapid AI progress and a hegemonic mindset seeking to maintain global technological dominance.

Gary Marcus, an American AI scholar, wrote in his own blog that Kimi K3 "is largely on a par with the best American models." The emergence of Chinese models "calls the business models of OpenAI and Anthropic into serious question, may kill or greatly undermine their IPOs. It also casts serious doubt on the never-too-plausible idea that America might 'win' the AI race."

Advocating open cooperation

Despite intensifying competition and governance tensions in AI technology, China has always advocated for the development of a global AI governance framework.

In October 2023, China released the Global AI Governance Initiative, outlining the principles of developing AI for good, ensuring security and controllability,and promoting inclusiveness and fairness. Calling for developing open, fair and efficient governing mechanisms, the initiative has provided systematic, practical and workable guidance for global AI governance.

In July 2025, China released the Action Plan for Global AI Governance, proposing six core principles of promoting AI for good and in service of humanity, respecting national sovereignty, aligning with development goals, ensuring safety and controllability, upholding fairness and inclusiveness, and fostering open cooperation.

China proposed the establishment of the World Artificial Intelligence Cooperation Organization in 2025, and one year later, the organization was formally established in Shanghai, with representatives from 29 countries signing the agreement as founding members.

During the 2026 World Artificial Intelligence Conference, China highlighted that all countries should uphold a people-centered approach and develop AI for the positive and for good, ensure that AI is an important driver for shared prosperity and common security, and join hands to build a just and equitable system for global AI governance.

"China has vigorously developed an open-source and open ecosystem, promoted global sharing of technological achievements and application experience, and firmly rejected closed and exclusive development models, as well as technological monopolies," said Zhou Haibing, deputy director of the National Development and Reform Commission, at a press conference in June, adding that China is committed to building an open, inclusive and mutually beneficial framework for international AI cooperation.