Beijing's advanced AI concerns (including Mythos/frontier models) fuel discussion of stricter China export controls on AI chips and technologies ahead of Trump-Xi meeting.
Beijing is growing anxious that Anthropic's Mythos could be used against China, adding a volatile new issue to tensions ahead of a planned summit between Xi Jinping and Donald Trump. Officials in Beijing are concerned about the cyber capabilities of Mythos and other frontier AI models developed by US companies, according to people familiar with the matter. They view these systems as potential offensive weapons and question why Anthropic is restricting Chinese access for normal purposes.
China wants to preserve a relatively warm atmosphere before Xi's planned September 24 visit to the United States and is seeking a "win-win" approach at AI talks expected before the leaders meet. However, suspicion remains high between the governments. The US has expressed concern that Chinese companies are using American frontier models to train their own versions. US Treasury Secretary Scott Bessent last month floated unspecified sanctions if model developers are found to have stolen intellectual property, following Beijing-based Moonshot AI's release of a system said to rival some of the best American models.
China's Commerce Ministry, which oversees export controls, warned last week that it would take "all necessary measures" in response to the US threats. According to people familiar with the matter, Beijing is not preparing any immediate action against Anthropic or other US AI developers, but officials are drawing up potential responses should the US take measures against Chinese companies in the sector. China's options could include imposing sanctions or adding American AI companies to one of the nation's restricted-entity lists.
Trump has signaled that his administration is weighing additional controls on AI while remaining wary of measures that could weaken US companies against Chinese competitors. "We have to be careful in both ways," he told reporters in the Oval Office. "We don't want to restrict them where all of a sudden, we come in second to China." White House spokeswoman Liz Huston said the administration was "doubling down on innovation to widen the gap between America and the rest of the world."
The tensions illustrate how AI is becoming an increasingly unpredictable front in the US-China relationship. The two countries remain strategic rivals despite reaching a 12-month truce last October over tariffs and supplies of rare earth materials. While that deal is widely expected to be extended, both sides continue to pursue self-sufficiency in key industries and maintain broad restrictions on cutting-edge technology. AI has increasingly been the focus of US national security curbs, most notably restrictions on China's ability to acquire cutting-edge Nvidia chips that power the technology. New limits on foreign-made robots and power inverters, industries dominated by China, have widened the dispute into sectors Beijing considers central to advanced manufacturing and economic growth.
Wu Xinbo, director of Fudan University's Center for American Studies in Shanghai, said China's concern about Anthropic fits a broader national security anxiety over US AI companies, citing their ability to collect customer data and share it with the US government or military. "That is why there is a growing concern over the national security challenge posed by those US AI companies," Wu said. "That's quite understandable. You have to be cautious."
Anthropic has emerged as a prominent advocate of tighter US restrictions on China's access to advanced AI. Chief Executive Dario Amodei has urged Washington to keep high-end chips out of China and require safety testing for powerful models. The company has also accused Chinese developers, including Alibaba's Qwen team, of improperly using outputs from its Claude systems to train competing models. Amodei warned last week that advanced Chinese AI could threaten US military superiority or deepen domestic repression, calling China's ruling Communist Party "the most capable threat" among authoritarian governments. Last year, Anthropic blocked its services from Chinese-controlled companies, citing US national security concerns.
Because Anthropic does not operate in China, any sanctions imposed by Beijing would probably have limited immediate commercial impact. Washington's own policies toward Anthropic have been unsettled. In June, Commerce Secretary Howard Lutnick ordered the company to obtain licenses before providing its frontier Mythos 5 and Fable 5 models to foreign entities anywhere in the world. Anthropic temporarily disabled the models before reaching a resolution with the administration. The company is also mired in a feud with the Trump administration over the use of its AI by the US military, stemming from the company's insistence on extra safeguards for its tools. Officials have banned the federal government from using the company's technology, a decision Anthropic is challenging in court.
Beijing demonstrated its ability to retaliate in June, when it placed export controls on MP Materials and USA Rare Earth. That came after the Pentagon added major Chinese firms, including Alibaba and BYD, to a list of companies it says support the Chinese military.
Top American and Chinese officials have raised points of friction between their countries even as they lay the groundwork for the September summit. Vice Premier He Lifeng expressed "serious concern" over recent US economic and trade restrictions in a phone call with Bessent and US Trade Representative Jamieson Greer. Summaries of the call from both sides made no mention of AI.
At the same time, Chinese open-weight models—which can be freely downloaded and modified—have given Beijing an opportunity to position itself internationally as a champion of broadly shared AI. China has promoted open models, offered technology and training to developing economies, and helped launch a 29-member cooperation group that could give Beijing greater influence over future AI standards. While some US developers, including OpenAI, offer open-weight models, much of the American industry's focus has been on proprietary systems that companies sell to customers while retaining control over underlying technology. Chinese companies, including DeepSeek, have released more affordable, open models that typically lag behind the most cutting-edge software from US rivals but nonetheless offer a compelling product for the price.
China has its own security concerns with AI released by Chinese companies. Officials have discussed with AI companies, including Alibaba, how to mitigate risks posed by their increasingly powerful models. A social-media account linked to China's state broadcaster and frequently used to signal official policy thinking proposed last week a tiered system for releasing future AI models, with access determined by the power and potential risks of their capabilities.