Cooperation Without Illusions: A Realistic Path for U.S.–China AI Safety

Chinese journalist Yi-Ling Liu

By EVWorld.com Si Editorial Team

The recent Politico interview with journalist Yi-Ling Liu reignited a familiar argument: the U.S.–China "AI race" is exaggerated, and America risks overreacting. Liu, who has spent years reporting from China's tech sector and diaspora communities, brings a lens shaped by deep familiarity with Chinese society and skepticism toward Silicon Valley's geopolitical narratives. That perspective is valuable, but it also risks understating the strategic realities of dealing with an opaque, authoritarian state.

Still, Liu is right about one thing: framing AI as a zero-sum contest pushes both countries toward reckless acceleration. The U.S. fears China will exploit any pause. China fears U.S. rules are a strategic brake. Neither fear is irrational. Neither system is transparent. And neither government has earned unqualified trust.

But cooperation does not require trust. It requires verification.

1. Verification over virtue

Instead of broad pledges, both countries could adopt shared, auditable safety benchmarks for frontier models: stress tests, red-team protocols, and incident reporting that can be independently checked by technical teams. This mirrors the logic of arms-control agreements: you do not trust motives, you verify behavior.

2. Reciprocity in small, reversible steps

No unilateral concessions. Each action, such as limited data-sharing, model-evaluation access, or joint catastrophic-risk research, should be paired with a reciprocal step. If one side backslides, commitments unwind automatically.

3. Track-two channels insulated from political theatrics

Scientists, labs, and civil-society groups can maintain continuity even when official relations sour. These channels already exist in climate science and public health. Extending them to AI safety creates a buffer against political swings in Washington or Beijing.

Liu’s interview reminds us that the loudest voices in the U.S. often use China as a rhetorical device, sometimes to justify deregulation, sometimes to demand escalation. But dismissing all concern as hype is its own form of wishful thinking. China’s political structure, industrial policy, and military-civil fusion strategy make blind trust impossible.

The point is not to trust China. Nor is it to trust America’s tech sector. The point is to build a system where cheating is costly, cooperation is beneficial, and verification is continuous.

Cooperation without illusions is still cooperation, and in the realm of frontier AI, it may be the only kind that stands a chance.

Views: 557