US and China discuss a warning system for AI threats

What the dialogue would cover

Bessent said the two countries need to move from opacity toward greater openness in their AI relationship, as they do in other areas of cross-border cooperation.

The discussions continued after Trump’s visit to Beijing in May. Since then, concern about AI-related risks to public safety has grown. Several incidents over the summer highlighted the gap between what advanced models can do and the industry’s defensive measures:

At the start of this month, Anthropic employees warned that AI could be moving toward a scenario in which humanity is destroyed. In recent weeks, leaders of major frontier-model laboratories have called for development to slow until safety standards exist. They included OpenAI’s Sam Altman, Anthropic’s Dario Amodei and Elon Musk.

Amodei has also called on governments to help build an international coordination system, including one involving the Communist Party of China.

The argument over slowing down

Trump has largely rejected calls for a slowdown, although lawmakers from both parties support regulation of the industry. His administration argues that the US government already has sufficient criminal-law and regulatory tools to control AI developers.

Nvidia CEO Jensen Huang takes a similar position. He has said companies are already subject to cybersecurity laws and liability for damage they cause. In Huang’s view, existing rules should be enforced first, rather than allowing discussion of catastrophic futures to exempt companies from requirements already on the books.

Trump has warned that slowing development could hand AI leadership to China. He has framed the stakes bluntly: “The winner in AI wins overall.” His former science adviser David Sachs previously said China was unlikely to join a global agreement on the pace of AI development.

Cooperation with a strategic condition

The proposed dialogue faces an obvious contradiction. The US is trying to constrain China’s AI development by restricting exports of advanced chips and the technology used to make them. Bessent said those restrictions were not discussed at meetings this weekend.

Thomas Wolf, co-founder of open AI platform Hugging Face, which Nvidia recently agreed to acquire, questioned whether the US can build international cooperation while openly seeking to widen its own advantage. Responding to Amodei’s essay, Wolf wrote on X that this approach looked like an unproductive start to negotiations.

I think that tension matters more than the creation of another diplomatic label. A warning channel can help governments share information about incidents and risks, but it cannot by itself reconcile cooperation with a race to limit the other side’s capabilities.

The talks also covered trade issues and, according to Bessent, were “very successful” ahead of Xi Jinping’s visit to Washington. What remains unclear is whether the two countries can discuss AI safety while leaving the central technology restrictions outside the room.