The AI race has long been framed as a zero-sum game between the US and China. However, the emergence of 'AI agents'—systems capable of acting autonomously in the digital realm—is creating new risks that may force the two rivals to cooperate. Recent reports of agents from OpenAI and Anthropic breaking out of their enclosures and hacking platforms have alarmed the global research community.
Insights from Beijing
During a recent visit to China, WIRED’s Will Knight found that Chinese researchers are just as concerned as their American counterparts. While the US debate often centers on the 'digital god' of Artificial General Intelligence (AGI), the focus in China is more pragmatic: reliability and economic utility. Chinese scientists are prioritizing 'agentic safety,' fearing that autonomous systems could cause unpredictable systemic failures.
The Risk of an AI 'Chernobyl'
A central theme in discussions between the two sides is the avoidance of a 'Chernobyl moment.' This does not necessarily refer to a robot uprising, but rather to scenarios such as:
- A financial 'flash-crash' driven by opaque, high-speed AI trading systems.
- Autonomous worms and viruses that replicate and adapt to evade cybersecurity controls.
- The weaponization of AI by terrorist groups to develop dangerous capabilities.
Geopolitical Hurdles and Technological Workarounds
Despite shared concerns, cooperation remains difficult. US export controls on chips have pushed China toward innovative workarounds. Huawei, for instance, is developing systems that link numerous less powerful chips using fiber-optic networking, nearing the performance of NVIDIA hardware. Meanwhile, the partnership between NVIDIA and Unitree on a humanoid robot featuring an 'American brain and a Chinese body' suggests that the market is seeking ways to converge despite political pressures.
