Sino-US AI Dialogue Outlook: Anthropic Model Sparks New Negotiations, Huawei Chip Plays Strongest Card

According to DeepTech Beat monitoring, China and the United States are about to launch a new round of AI security dialogue. The direct catalyst is the cutting-edge model Mythos from the American company Anthropic, which can autonomously infiltrate government databases, banks, and hospital networks, and is seen by the U.S. as an "unprecedented cyber weapon." Although Mythos is made in the U.S., it has demonstrated that this destructive capability is technically feasible. Once a similar ability is mastered by adversaries or hacker organizations, neither China nor the U.S. can withstand it. Meanwhile, DeepSeek has announced for the first time that its new model is compatible with Huawei chips, signaling that Chinese AI is reducing its reliance on the U.S. chip giant NVIDIA, adding a bargaining chip for China. U.S. senior officials have confirmed that the White House hopes to open a communication channel through the recent meeting between the Chinese and American leaders.

The Biden administration has twice pushed for China-U.S. AI communication. At the 2024 Geneva first-round technical negotiations, according to four attendees, China regarded the U.S.'s concern about "AI out-of-control" as an academic hypothesis and instead pressured for a chip export ban, derailing the talks. In November of the same year, the two sides reached a limited agreement in Peru, committing not to allow AI to control nuclear weapon launches. Then-U.S. National Security Advisor Sullivan called the Peru agreement a "breakthrough," proving that China and the U.S. could achieve substantive outcomes in AI discussions. Therefore, he strongly recommended the new team to continue the negotiations, but this was not adopted. It was not until the emergence of Mythos that U.S. Treasury Secretary Benson proposed a federal review of future new models after hearing security reports from multiple banks, leading to a complete shift in the White House's stance.

However, the more fundamental obstacle is that the two countries are running in completely different races. Silicon Valley giants are striving for "Artificial General Intelligence" (AGI, AI with human-level general intelligence), believing that whoever achieves AGI first will gain irreversible technological dominance. China is not pursuing the smartest AI but is embedding existing AI into factories, hospitals, autonomous vehicles, and government systems to immediately create economic value. CSIS senior advisor Kennedy said, "China believes there is not a single race but multiple parallel races." When the two sides sit down together, the U.S. wants to discuss "what to do if superintelligence goes out of control," while China wants to discuss "why you prevent me from buying chips."

A recent Stanford report stated that the performance gap between Chinese and American models has "essentially closed," but the U.S. still holds the advantage: U.S. tech companies' AI investment is ten times that of their Chinese counterparts, and the number of data centers (the physical carriers of AI computing power) is also ten times greater. Industry experts suggest starting by tackling specific risks one by one, similar to the Peru nuclear weapons agreement. However, Kyle Chan from the Brookings Institution warns that mutual distrust is fueling an "all-or-nothing competition" in the security field, with both sides reducing security investments to run faster. Sullivan said that China's leadership theoretically recognizes the need for cooperation but their sense of urgency is much lower than that of the U.S. "However, their sense of urgency has now increased."