On July 20th, according to sources familiar with the matter, reported by the American news website Axios, the US government is considering banning Chinese artificial intelligence models.
The Trump administration has previously attempted several times to impose de facto bans on foreign open-source models. The rise of the Chinese open-source model, Kimi K3, last week has reignited these efforts. Those who advocate for free competition are deeply concerned about the possible consequences of this move.
A source close to the government told Axios that the U.S. Department of Commerce considered listing several Chinese AI laboratories on a “entity list” last year. The source said that the National Security Agency (NSA) and the Office of the National Cyber Coordinator at the White House also considered issuing a notice regarding the threats posed by Chinese AI laboratories last year, in an attempt to dissuade U.S. companies from using such technologies. The source added that the White House also considered issuing an executive order requiring U.S. companies to use Chinese models only if they can ensure the security of the technology and are responsible for any data breaches that may occur.
Another source close to the government revealed that the U.S. Department of Commerce also circulated draft rules last summer within the government, which targeted Chinese open-source models using supply chain security authorities. However, government officials who feared regulation would stifle innovation stopped this initiative.
However, the situation has changed now.
On July 16th, Chinese technology company Yuezhi Dark Side released the world's first open-source model, Kimi K3, which has 2.8 trillion parameters. The company stated that this model surpasses all competitors, except for Claude Fable 5 under Anthropic and OpenAI GPT-5.6, in overall performance.
Information revealed by the dark side of the moon shows that the Kimi K3 model natively supports image input and a context of 1 million Tokens. It employs the Kimi Delta Attention (KDA) system, attention residuals, and a highly sparse hybrid expert architecture. When processing each Token, only 16 experts from 896 experts are effectively activated to improve the efficiency of computing power utilization.
The dark side of the moon states that the related architecture and data training methods have increased the overall scalability efficiency by approximately 2.5 times compared to Kimi K2. It also mentioned that in internal evaluations, Kimi K3 outperforms GPT-5.5 and Claude Opus 4.8 in terms of end-to-end knowledge processing capabilities.
Bloomberg and other US media pointed out that this means the gap between China’s AI and the most advanced models in the United States may have been narrowed to a few months.
China's technological strength has increased, and with the departure of key pro-market figures such as Sriram Krishnan, a former senior policy advisor on artificial intelligence at the White House, momentum within the U.S. government to impose such bans is resurging.
A source familiar with government discussions said, “What actually happens is often slower, but it lasts longer.” Another insider described the current strategy as not outright prohibition, but rather attempting to pressure China by damaging its reputation, emphasizing the potential risks and security vulnerabilities behind it, as well as the governance issues that arise from these issues.
As of now, neither the White House nor the Department of Commerce have responded to requests for comment.
But Axios also understands that the real problem is that, compared to cheap and practical Chinese models, there is currently no strong alternative in the United States.
Previously, American media described the performance improvements brought by the Kimi K3 as shocking to some AI observers and investors, leading to market fluctuations similar to those experienced when DeepSeek released its results last year. The release of the Kimi K3 has once again raised doubts about the huge returns from AI investments in Silicon Valley, and has shaken the confidence that the United States has maintained a leading position in AI for a long time.
Axios pointed out that China has gained more confidence in the AI competition. Whether the technology is open-source or not, the pressure on the American industry to take measures is increasing.
Sources close to the U.S. government said that although the measures under consideration are not outright bans, they will still have a ‘stifling effect’ on U.S. companies that rely on Chinese open-source technologies. The sources revealed that every 3 to 5 months, leading AI laboratories or their allies will propose to the government the prohibition of open-source models.
White House AI and cryptocurrency advisor David Sacks wrote on the X platform on the 19th: "We are at a critical turning point in AI policy. Leading closed-source laboratories, which already have a duopoly in terms of model revenue, are hoping that the government will eliminate their open-source competitors." "They have exposed their hand. It's now time for other companies in Silicon Valley—most of whom still believe in open competition—to respond."
