Could a US-China 'hot line' be set up for AI risks?

With unchecked development of artificial intelligence sparking deep fear even among leaders in the sector, the United States is looking at establishing a dedicated communication channel with China on the issue.

The link was suggested during talks on the weekend between senior US and Chinese officials, ahead of a summit later this week between US President Donald Trump and Chinese counterpart Xi Jinping.

"The US has proposed that we have a notification mechanism between the two countries" on AI, US Treasury Secretary Scott Bessent said after his talks on Sunday with Chinese Vice Premier He Lifeng.

The goal was to reach "a shared vision of common goals and common threats", he said, without providing further details.

Trump has publicly dismissed concerns voiced by some in the AI sector that humanity could be wiped out by highly advanced AI systems within a decade if adequate safeguards are not put in place.

Fears have risen with reports that several models from Anthropic and OpenAI broke out of confined environments to access the internet.

Work towards "recursive superintelligence", where AI models build future versions of themselves in ways that humans may not be able to grasp, have sharpened the anxiety, as have moves towards AI controlling unmanned weapons.

The United States and China, engaged in competition for leadership in AI development, are reluctant to impose significant restrictions on domestic companies for fear of losing an advantage.

"We are seeing both sides starting to realise that there are possibilities of an incident occurring and of misuse, to the detriment of both countries," said Jean-Marc Rickli, head of global and emerging risks at the Geneva Centre for Security.

He drew a parallel with the US proposal for a communication channel with China on AI and the Cold War "hot line", which was established after the 1962 Cuban missile crisis.

- 'Difference in views' -

AI has not yet triggered alarm on that scale. But Rickli noted a CNN report last week that said the United States military almost boarded a Chinese ship on the basis of an incorrect report drawn up using AI.

"If the United States and China ... are going to safely navigate this brave new world, they will need the ability to communicate quickly and effectively about an emerging AI crisis," said Lucy Luo, of the Washington-based Carnegie Endowment for International Peace think tank.

Yet the priorities and interests of the two superpowers are not always aligned.

A French researcher, Mathilde Viellet of the French Institute of International Relations (IFRI), noted that, since 2023, there has been a non-governmental dialogue between experts and research centres on the world's most advanced AI models, including those from OpenAI, Anthropic, DeepSeek and Alibaba.

And there, both sides show "a difference in views on safety", she said.

"The Chinese say that they have been working for several years on things such as deepfakes, information security, social disruption" while the Americans "are mostly worried about a loss of control and 'misalignment' of AI, when it acts in an autonomous way".

The US consultancy Teneo said that "neither government is yet willing to commit to binding constraints on AI development".

There is also the US preference for the development of proprietary AI models, owned by just one company, whereas China leans more towards open source models, for accelerated development and scale.

Rickli said the challenge, for Washington and Beijing, will be to find "the smallest common denominator", which might involve coordination on crisis response or "possible 'misuse' by third actors".

He said he believed "a hot line makes sense" in the military sphere, where AI is increasingly being used for targeting and accelerated decision-making.

"But is that realistic, when operational speed is a key criterion for strategic advantage?" he asked.

- Communication options -

For IFRI's Viellet, there are two issues on which Washington and Beijing have a shared interest in communicating: "Loss of control (of AI) and its use by bad actors."

She said three paths were available to them.

One would be a hot line to use if an AI model started acting in a harmful or unexpected way.

Another would be as a mechanism to divulge the malicious use of AI -- such as Anthropic's revelation that its Claude model had been used in a way that could go towards developing biological weapons.

The problem on that point, though, could be defining the perimeter of what that might cover.

"If cyber groups close to the Russians used AI to attack US interests, would the Chinese have an interest in sharing that with the Americans?" Viellet asked.

The third path could be sharing information on the evaluation of the safety of AI models, she said.

Sam Altman, CEO of OpenAI, last week told Fortune magazine: "Even if just the US and China could agree on some shared standards and testing for development of this technology, I think that'd be a wonderful accomplishment."

Such an achievement would merit Trump and Xi earning the Nobel Peace Prize, Altman said.

But, for Matt Sheehan, at the Carnegie Endowment for International Peace, "the most important actions on AI governance and safety will be domestic actions that the US and China take -- not international or bilateral agreements".

fz/sva/rmb/

Alibaba