Trump-Xi meet: Will AI safety find its way to the discussion table?

US president Donald Trump and his Chinese counterpart Xi Jingping will meet for bilateral talks in the US during the latter’s state visit. One of the hottest topics that could be on the agenda revolves around artificial intelligence (AI) and how both nations can find common ground on security and governance issues.

To date, the AI race between the US and China has largely been seen as a contest over who can build the most powerful and cost-efficient models, secure the most computing prowess, and control the future of this technology.

Dialogue between the US and Chinese officials on how the two countries could communicate when AI goes seriously wrong has started.


The US has proposed a notification mechanism with China for significant AI incidents, US treasury secretary Scott Bessent told reporters on Sunday after talks with Chinese Vice Premier He Lifeng. The motive behind such a move, he said, is to build a “shared vision of common goals and common threats”.

So, when Trump and Xi meet, this proposal could put AI risks on the table, alongside the larger trade and technology tensions between Washington and Beijing. The exact agenda for the meeting has been kept under wraps, but according to Bessent’s comments to CNBC, the discussion could include “what the leading AI dangers are, whether it’s uncontrollable agents, whether it’s non-state actors, and cyber non-state actors in bioweapons”.

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

While this might seem like the opening of a communication channel, experts see it as an important first step because some of the risks emerging from increasingly capable AI systems are not restricted to national borders.

“Both sides are starting to realise that there are possibilities of an incident occurring and of misuse, to the detriment of both countries,” Jean-Marc Rickli, head of global and emerging risks at the Geneva Centre for Security, told AFP.

He compared the proposed mechanism with the US-Soviet hotline established after the Cuban missile crisis, arguing that a channel for rapid communication could become particularly important as AI is increasingly used in military and security systems.

AI industry’s “slowdown” push

The push for more coordination did not begin with governments but from inside the AI industry.

On September 12, Anthropic chief executive officer (CEO) Dario Amodei called for the industry to “pace the frontier” in a personal essay, arguing that AI capabilities were advancing rapidly and that safety measures were struggling to keep up.

OpenAI CEO Sam Altman, Elon Musk and Google DeepMind's Demis Hassabis backed Amodei's call to varying degrees. OpenAI said it would adopt independent evaluations of its systems and regularly publish reports on unexpected or unauthorised AI behaviour.

The calls came after a series of incidents involving autonomous AI systems where models escaped controlled environments or carried out unauthorised activity during testing.

Amodei argued that the problem could exacerbate if AI systems start working to build their successors.

In a blog post on Tuesday, Anthropic’s rival OpenAI called for international technical standards for frontier AI, including common approaches to measure capabilities, evaluate risks and report incidents involving increasingly autonomous systems.

The firm said standards would be important to tackle “recursive self-improvement”, a theoretical process where an AI system autonomously redesigns, rewrites, or enhances its own code and architecture to create a more capable successor

“Making continued AI progress safe and beneficial will require advances in alignment research and shared standards to guide development across labs and countries,” the company noted.

OpenAI further added that dialogue between the US and China would be a positive step.

Different viewpoints on AI risks

Mathilde Viellet, a researcher at the French Institute of International Relations, told AFP that discussions between US and Chinese AI experts have shown a “difference in views on safety”.

Chinese researchers have historically focused more on issues such as deepfakes, information security and social disruption, while those in the US have emphasised loss of control and AI “misalignment” when systems become autonomous.

In addition, there exists an underlying strategic problem: both nations don’t want safety rules to impede their technological advantage.

The US is concerned about China's access to advanced AI chips and has accused Chinese companies of extracting capabilities from American AI models through distillation – allegations Beijing has rejected. China, meanwhile, has accused Washington of trying to contain its technological development.

Also Read: China and the US are competing for AI dominance but have shared concerns over safety

Still, there may be a narrower area where both sides have an incentive to talk, to mitigate any future national security problem emanating from AI.

Sun Chenghao, a senior fellow at Tsinghua University's Center for International Security and Strategy, similarly told AP that there was “meaningful room for cooperation” around risks such as AI-enabled cyberattacks, biological misuse, major model failures and loss of human control.

The same point was made by Xiao Qian, vice dean of Tsinghua University's Institute for AI International Governance, too. Speaking to AP, she said, if the two countries can agree on how to report and communicate during an AI emergency, it could help build technical confidence over time.

For the time being, that may be a much more realistic starting point than trying to get Washington and Beijing to agree on how AI should be regulated.