AI Illustration: Chen Xia/GT

A telling juxtaposition recently emerged in China-US AI relations. In mid-October, the US Department of Homeland Security publicly ordered federal civilian agencies to remove AI technology developed by China's DeepSeek from their information technology systems, prompting China's Foreign Ministry to reiterate its opposition to the overstretching of the concept of national security. 

Meanwhile, former US treasury secretary Robert Rubin called at the Bund Summit in Shanghai for China and the US to put aside their mutual distrust to address the risks posed by AI, and learn from Cold War-era arms control.

This juxtaposition raises a question: If Washington approaches AI development through the lens of hegemonic preservation, insisting on treating China's technological progress as a threat to US survival that must be eliminated, how much room will remain for cooperation on shared security?

Rubin's analogy deserves serious consideration. One important prerequisite for Cold War arms control was a shared understanding between the US and the Soviet Union that, regardless of their mutual hostility, they had a common interest in preventing human annihilation. Although AI is a general-purpose technology and its governance cannot simply replicate the management of nuclear weapons, the principles of recognizing each other's legitimate security interests and accepting reciprocal constraints remain applicable. 

The current direction of US policy is troubling. When the US released its AI Action Plan, it explicitly emphasized consolidating US leadership in the field. When this pursuit becomes absolute, it gives rise to a dangerous logic: The US can be secure only if its advantages remain unchallengeable; China's narrowing of the technological gap is itself a threat; and the priority in China-US competition becomes the US doing whatever it takes to trip China up, rather than working with China toward common goals.

This is how "hegemonic anxiety" manifests itself in AI governance. It equates a potential narrowing of technological advantages with an actual infringement of national security interests, while treating a competitor's room for development as something that can be curtailed or even trampled on at will. Along this path, the language of "arms control" could also be used to package unilateral constraints and entrench existing technological disparities in international rules.

When China's sovereignty, security and development interests face tangible constraints, it will inevitably strengthen its capacity for independent innovation and self-defense. Washington may then interpret these efforts as a "new threat" and impose further restrictions. 

The US needs to accept reality: China will not relinquish its right to develop advanced technologies, and unilateral bullying cannot deliver lasting stability. Recognizing this is a sign of strategic maturity. A constructive relationship of strategic stability requires both sides to be able to compete while cooperating to address shared dangers. 

Technological developments are making this shift increasingly urgent. As AI agents capable of using tools and executing tasks enter more areas of application, risks are expanding beyond erroneous answers to include unauthorized actions. OpenAI disclosed that during internal cybersecurity evaluations in July 2026, its models circumvented controls designed to isolate them from the internet and compromised parts of OpenAI's internal research infrastructure and Hugging Face's systems.

Such risks do not disappear depending on which country's company name appears on a model. If the US seeks to seal off security knowledge behind competitive barriers in pursuit of hegemonic advantage, the US itself and the wider world may ultimately pay a higher price. China and the US could begin by establishing clearly defined, mutually applicable cooperation arrangements covering the incident reporting, the tests for AI agents exceeding their authorized permissions, the emergency communication channels and the human control over critical systems.

China has already laid the groundwork for cooperation. The China-proposed Global AI Governance Action Plan, released in 2025, proposes measures including the risk testing and evaluation system, the sharing of information as well as the development of emergency response of AI safety risks and threats. The goodwill conveyed by these proposals deserves a practical response from the US. 

The author is director of the International Research Institute of Global Cyberspace Governance at Fudan University. opinion@globaltimes.com.cn