China is increasingly concerned about the Mythos AI model ahead of the Trump-Xi summit - Bloomberg
Kyiv • UNN
Beijing is worried about the cyber capabilities of Mythos and other American AI models, viewing them as potential offensive weapons. China is developing possible responses, including sanctions, if the US takes action against its companies in this sector.
China is increasingly concerned that Mythos from Anthropic PBC could be used against the world's second-largest economy, adding a new sharp issue to the already heightened tensions ahead of the planned summit between Chinese leader Xi Jinping and U.S. President Donald Trump, Bloomberg reports, writes UNN.
Details
According to people familiar with the matter, officials in Beijing are worried about the cyber capabilities of Mythos and other advanced artificial intelligence models developed by American companies. They see the potential for using models like Mythos as offensive weapons and question why Anthropic denies China access to the systems for ordinary purposes, the sources said.
China wants to maintain a relatively warm atmosphere ahead of Xi's planned visit to the U.S. on September 24 and seeks a "mutually beneficial" approach in the artificial intelligence talks expected before the leaders' meeting, the sources said. Despite this, suspicions between the governments remain high, and the U.S. expresses concerns that Chinese companies are using American advanced models to train their own versions.
Alibaba joined ranks of Chinese AI breakthroughs with new Qwen model03.08.26, 10:54 • 4092 views
U.S. Treasury Secretary Scott Bessent last month imposed unspecified sanctions if model developers detect theft of intellectual property after Beijing-based company Moonshot AI released a system that is claimed to be able to compete with some of the best American models. China's Ministry of Commerce, which controls export controls, warned last week that it would take "all necessary measures" in response to U.S. threats.
Individuals familiar with the matter said that Beijing is not preparing any immediate actions against Anthropic or other American AI developers, but officials are developing possible responses if the U.S. takes action against Chinese companies in this sector. China's options could include imposing sanctions or adding American AI companies to one of the national lists of restricted organizations, these sources said.
Trump has made clear that his administration is considering additional measures to control artificial intelligence, while remaining cautious about measures that could weaken American companies against Chinese competitors. "We have to be careful in both directions," he told reporters in the Oval Office last Wednesday. "We don’t want to restrict them where all of a sudden, we come in second to China."
White House spokesperson Liz Huston said the administration is "doubling down on innovation to increase the gap between America and the rest of the world."
This tension, the publication writes, illustrates how AI is becoming an increasingly unpredictable front in U.S.-China relations. The two countries remain strategic rivals, despite reaching a 12-month truce in October last year on tariffs and supplies of rare earth materials. Although this agreement is expected to be extended, both sides continue to seek self-sufficiency in key industries and maintain broad restrictions on advanced technologies.
AI is increasingly becoming a subject of U.S. national security restrictions, including limits on China's ability to buy advanced chips from Nvidia Corp., which power this technology. New restrictions on foreign-made robots and power inverters, sectors where China dominates, have extended the dispute to areas that Beijing considers central to advanced manufacturing and economic growth.
Wu Xinbo, director of the Center for American Studies at Fudan University in Shanghai, said China's concerns about Anthropic align with broader concerns about the national security of American AI companies, citing their ability to collect customer data and share it with the U.S. government or military.
"That's why there is growing concern about the national security challenge posed by these American AI companies," Wu said. "It's quite understandable. You have to be careful."
Anthropic has become a prominent advocate for stricter US restrictions on China's access to advanced artificial intelligence. CEO Dario Amodei has called on Washington to prevent high-quality chips from reaching China and to require safety tests for powerful models. Separately, the company accused Chinese developers, including the Qwen team at Alibaba Group Holding Ltd., of improperly using the results of its Claude systems' work to train competing models.
Anthropic's chief also warned last week that advanced Chinese artificial intelligence could threaten US military superiority or deepen internal repression, calling the ruling Communist Party of China "the most powerful threat" among authoritarian governments. Last year, the company blocked its services for companies controlled by China, citing US national security concerns. Since Anthropic does not operate in China, any sanctions imposed by Beijing would likely have limited immediate commercial impact.
Washington's own policy toward Anthropic remains uncertain, the publication writes. In June, US Commerce Secretary Howard Lutnick ordered the company to obtain licenses before providing its advanced models Mythos 5 and Fable 5 to foreign organizations anywhere in the world. Anthropic temporarily disabled these models before reaching a settlement with the administration.
The US lifts restrictions on powerful AI models, Anthropic reported01.07.26, 09:57 • 3596 views
The company has also been embroiled in a feud with the Trump administration over the use of its artificial intelligence by the US military, triggered by the company's insistence on additional safety measures for its tools. Officials banned the federal government from using the company's technologies, a decision Anthropic is challenging in court.
Beijing demonstrated its ability to retaliate against US actions against its tech companies in June, when it imposed export controls on MP Materials Corp. and USA Rare Earth Inc. This came after the Pentagon added major Chinese firms, including Alibaba and BYD Co., to a list of companies it says support China's military.
Senior US and Chinese officials have raised issues of friction between their countries, even as they prepared for a September summit. Chinese Vice Premier He Lifeng expressed "serious concern" over recent US economic and trade restrictions during a phone call with Bessent and US Trade Representative Jamieson Greer, without specifying which measures were taken. Brief reports on the call from both sides did not mention artificial intelligence.
At the same time, Chinese open-weight models (models with open weights), which can be freely downloaded and modified, have given Beijing an opportunity to assert itself internationally as a leader in the widespread use of artificial intelligence. It has promoted open models, offered technology and training to developing countries, and helped launch a collaboration group with 29 participants that could give Beijing greater influence over future standards.
While some American developers, including OpenAI, offer models with open weights, a significant portion of the US industry focuses on proprietary systems that companies sell to customers while maintaining control over the underlying technology. Chinese companies, including DeepSeek, have released more accessible, open models that typically lag behind the state-of-the-art software of American competitors but nonetheless offer an attractive product for their price.
China has its own security concerns related to artificial intelligence released by Chinese companies. Officials have discussed with AI companies, including Alibaba, how to reduce risks arising from their increasingly powerful models.
The Yuyuantantian social media account, associated with China's state broadcaster and often used to showcase official political positions, last week proposed a multi-tiered release system for future AI models, with access to be determined by the power and potential risks of their capabilities.