US and China Seek Safer Ground in AI Competition

The diplomatic window for managing artificial intelligence risks between the United States and China is narrowing. Even as Treasury Secretary Scott Bessent prepares to meet Chinese Vice Premier He Lifeng, President Donald Trump has dismissed public warnings about runaway AI as a hoax. Simultaneously, state media in China has accused American firms of using safety concerns to slow down Chinese progress. This creates a volatile environment where neither side is willing to slow down, yet both recognize that the technology will not wait for diplomacy to catch up.

According to analysis by GN technics/ai (en-US), the solution lies in avoiding two common diplomatic traps. Setting goals too low results in meaningless photo opportunities and vague statements about innovation. Setting goals too high leads to failure because neither country has the trust or capacity to enforce sweeping restrictions. The recommended path is for both nations to pursue AI safety in parallel, focusing on technical exchanges that make their respective systems safer on their own terms.

Technical exchanges drive self-interested safety

The proposed approach relies on self-interest rather than trust. Both countries benefit if advanced AI models remain controllable, regardless of where they are built. This means sharing best practices on how to test models for dangerous capabilities, such as the ability to help amateurs develop bioweapons or evade human oversight. American companies have studied these risks for years, and similar safety standards are emerging in China. Exchanging insights on mitigating these specific hazards could make systems in both countries meaningfully safer without requiring them to surrender competitive advantages.

This strategy is less about negotiating a sweeping arms-control treaty and more about establishing shared safety procedures for a technology that neither government fully understands. It does not require identical conceptions of risk or strict reciprocity. Instead, it acknowledges that if one country loses control of a powerful system, the other suffers the consequences. This practical framework allows for progress even in an atmosphere of deep mutual suspicion.

Crisis protocols prevent accidental escalation

To make this safety framework work, it must be paired with a government-to-government crisis communication protocol. The stakes are high because AI systems can act autonomously or malfunction in ways that mimic deliberate aggression. If an American model hacks Chinese networks, or a Chinese drone attacks a US Navy vessel, immediate clarity is needed to determine if the incident was accidental or authorized. Without a clear channel for rapid communication, misunderstandings could spiral into international crises.

Avoiding the trap of failed diplomacy

The catch in this approach is that it requires sustained technical engagement rather than political grandstanding. It demands that experts from both sides work together on specific testing and monitoring methods. If the talks revert to high-level rhetoric or fail to address the specific technical mechanisms of safety, they will fail to meet the actual stakes of the moment. The goal is not to stop the AI race, but to install guardrails that prevent catastrophic accidents while allowing competition to continue.