US and China Face Complex Path to AI Safety Cooperation

Global efforts to manage the risks of artificial intelligence are reaching a critical juncture, relying heavily on cooperation between the United States and China. However, these two superpowers currently view each other as primary obstacles rather than partners in safety. As tensions rise, the prospect of a unified global approach appears increasingly fragile, with both nations prioritizing their own competitive advantages over shared security protocols.

A high-level meeting is scheduled for next week in Washington, where President Donald Trump and Chinese leader Xi Jinping are expected to address AI governance. The urgency is driven by warnings from American tech leaders who argue that preventing catastrophic AI failures requires some level of engagement with Beijing. Yet, the political atmosphere remains hostile, with Washington arguing that strict regulation would only benefit Chinese competitors, making any potential agreements likely to be superficial.

Divergent strategies create friction

The two countries are pursuing AI development through fundamentally different methods. The United States focuses on frontier capabilities powered by advanced chips and expanding data centers, while China emphasizes widespread global adoption of affordable, open-source models. This divergence has led to accusations on both sides; the US claims China is attempting to extract capabilities from American systems, while Beijing argues the US is trying to establish an unfair monopoly.

Trust is a significant barrier to any deal. David Sacks, a top AI adviser to the White House, has stated that unlike nuclear arms control during the Cold War, it is impossible to verify if China honors AI commitments. He argues that the lack of trust and verification mechanisms makes deep cooperation dangerous, suggesting that the US should not concede ground to Beijing.

Industry calls for cautious engagement

Leading American AI companies hold mixed positions on this issue. Firms like Anthropic and OpenAI restrict access within China, citing national security concerns. However, they also advocate for some form of cooperation to manage safety risks. Dario Amodei, CEO of Anthropic, recently argued that while the US should maintain restrictions on advanced chip sales to China, a complete lack of dialogue is not the solution.

According to reporting from GN technics/ai (en-US), experts suggest that formal treaties may be unattainable in the current climate. Instead, the goal should be to create political space by acknowledging that AI poses risks to both nations. This approach allows for limited, practical steps without requiring a full consensus on regulatory frameworks, offering a potential starting point for future engagement.

Verification remains a major hurdle

The core challenge is not just intent, but the ability to monitor compliance. Without a trusted mechanism to verify that both sides are adhering to safety standards, any agreement remains theoretical. This technical and political gap means that even if leaders agree in principle, the practical implementation will face severe scrutiny and skepticism from domestic stakeholders in both countries.