A Look Ahead at the US-China AI Dialogue: The Anthropic Model Forces New Negotiations, Huawei's Chips Offer Its Strongest Trump Card

According to Beating's monitoring, a new round of AI security dialogue between China and the US is about to begin. The direct catalyst is the cutting-edge model Mythos from the US company Anthropic, which can autonomously infiltrate government databases, bank networks, and hospital networks, and is considered by the US as an "unprecedented cyber weapon." Although Mythos is American-made, it proves that such destructive power is technically feasible, and if similar capabilities fall into the hands of adversaries or hacker groups, neither China nor the US could withstand the consequences. Meanwhile, DeepSeek announced for the first time that its new model is compatible with Huawei chips, indicating that Chinese AI is reducing its dependence on US chip giant Nvidia, adding to China's negotiating leverage. Senior US officials confirmed that the White House hopes to use the upcoming meeting between Chinese and US leaders as a window to open communication channels. The Biden administration has twice promoted Sino-US AI communication. During the first round of technical negotiations in Geneva in 2024, according to four participants, China viewed the US's concerns about "AI going out of control" as academic hypotheses and instead pressured for a chip export ban, causing the talks to derail. In November of the same year, the two sides reached a limited agreement in Peru, promising not to allow AI to control nuclear weapons launches. Then-US National Security Advisor Sullivan called the Peru agreement "the seal was broken," proving that the US and China could achieve substantial results in AI. Therefore, he strongly recommended that the new team continue this effort during the handover, but this was not adopted. It wasn't until Mythos emerged, and US Treasury Secretary Bessant, after receiving security briefings from several banks, proposed federal review of future models, that the White House's attitude completely shifted. However, the more fundamental obstacle was that the two countries were running completely different races. Silicon Valley giants were racing towards "Artificial General Intelligence" (AGI, meaning AI with human-level comprehensive intelligence), believing that whoever created AGI first would gain irreversible technological hegemony. China, on the other hand, wasn't pursuing the smartest AI, but rather embedding existing AI into factories, hospitals, driverless cars, and government systems, allowing it to immediately create economic value. CSIS senior advisor Kennedy said, "China believes there is no single competition, but multiple parallel competitions." When the two sides sat down together, the US wanted to discuss "what to do if superintelligence goes out of control," while China wanted to discuss "why you won't let me buy chips." A recent Stanford report states that the performance gap between Chinese and American models has "effectively closed," but the US still holds an advantage: US tech companies invest 10 times more in AI than their Chinese counterparts, and the number of data centers (the physical carriers of AI computing power) is also 10 times greater. Industry experts suggest addressing specific risks one by one, similar to the Peruvian nuclear arms deal. However, Kyle Chan of the Brookings Institution warns that mutual distrust is fueling a "race to the bottom" in security, with both sides cutting security spending to move faster. Sullivan says that while the Chinese leadership acknowledges the necessity of cooperation in principle, their sense of urgency is far lower than that of the US. "However, their sense of urgency has increased."