US officials are pitching a new AI safety notification mechanism with China as a modest step toward reducing global risk, but experts say the effort is being overshadowed by Donald Trump’s broader “win the AI race” posture and by a lack of technical expertise in the proposed channel. With Trump set to meet Xi Jinping on Thursday and Friday, the talks could shape whether Washington and Beijing treat AI safety as a shared problem or just another front in strategic competition.
A proposed AI safety channel, but little trust
Treasury Secretary Scott Bessent said both sides had already discussed creating a way for each country to send alerts when its AI systems pose threats or behave unpredictably. In theory, the channel could also be used to share a “vision of common goals.”
But the idea arrives amid escalating tensions. Trump has tightened export controls aimed at limiting China’s access to advanced technology, Congress is weighing more restrictions on chips and hardware, and the Justice Department recently accused six Chinese AI companies of stealing from US frontier labs. China has strongly disputed those claims, and it has not publicly acknowledged the proposed alert system.
Why experts say the plan may be too shallow
People familiar with the plan told Politico that the mechanism appears superficial because it would not involve officials with technical expertise who could assess whether an alert describes a real and urgent risk. One source dismissed it bluntly, saying “the scientific, technical term is BS.”
Sam Bresnick, a research fellow at Georgetown University’s Center for Security and Emerging Technology, said the existence of a channel is better than nothing, but he does not expect it to move the needle much. He argued that including technical experts on both sides would allow deeper discussion of how each country evaluates risk, especially when military concerns are at stake.
What the experts want instead
- Risk classifications both sides can agree on
- Clear notification triggers
- Response protocols for urgent warnings
- Regular review procedures to keep the system credible
Su Lian Jye, chief analyst at Omdia, said the alerts could improve transparency, but they would not erase distrust over issues such as distillation, pricing competition, and the pace of AI development. He said turning the proposal into something meaningful would require much broader cooperation.
Trump’s China rivalry shapes the AI debate
Critics say Trump’s framing of AI as a contest to be won may undercut the very cooperation needed to manage the technology’s risks. On Truth Social, Trump declared, “WHOEVER WINS AI, WINS! We are leading China, and all others, and will continue to do so.”
That rhetoric sits uneasily beside the idea of an open safety line between the two countries. Bresnick said China may see “AI safety” as code for attempts to suppress Chinese AI capabilities rather than a genuine effort to coordinate on shared dangers.
At the same time, China appears willing to keep talking. Even if some US-China experts expected Beijing to walk away, China showed up for the meetings and has another discussion scheduled for November.
What’s actually at stake in US-China AI talks
Alvin Wang Graylin, a global technology strategist and digital fellow at Stanford Digital Economy Lab, said the US and China face overlapping risks, including misuse by hackers, terrorists, and other bad actors. He also argued that China already has more experience auditing AI systems, while the US has only recently begun serious auditing efforts.
Graylin said he was encouraged that the talks produced something concrete, even if only an initial notification idea. In his view, the follow-up meeting suggests both countries see value in continuing the conversation.
Beyond “winning” a frontier race
Graylin has argued that the US is trapped in a “deeply flawed” narrative about an AI race that cannot deliver a permanent advantage. He wrote in a September article for the Quincy Institute that capability gaps at the frontier often last only a few months and that there is no lasting moat that would secure a durable lead.
He also warned that there are really two races: one for intelligence and one for market dominance. By his account, US firms are still making more money at the frontier, but Chinese open models are gaining users globally. He predicted some could soon exceed 80 percent market share, especially among businesses looking for lower-cost alternatives that may better protect privacy.
The bigger problem: safety, not supremacy
Graylin’s central argument is that the US and China are focusing too much on hypothetical supremacy while missing nearer-term dangers. He said the most dangerous systems may be smaller specialized models that could be used by rogue actors to develop biological, chemical, or cyber weapons.
He wrote that assuming US-China conflict blocks the kind of global cooperation needed to address those threats. In his view, if Washington wants safer deployment, it should broaden AI safety conversations, limit export controls to genuine military and CBRNE threats, and consider more balanced domestic regulation.
“We cannot manage the multifaceted risks of the most powerful technology in human history while refusing to talk to the other nation building it,” Graylin wrote. He added that there is “no way to compete yourself to a safer world.”
Where the Trump-Xi meeting leaves things
The immediate outcome of the summit may be less important than the signal it sends. If the AI alert mechanism remains technical in name only, it could become just another symbolic gesture in a relationship defined by suspicion. If both sides invest in expert-level engagement, it might become the start of something more durable.
For now, though, experts say Trump’s rhetoric about domination makes the process harder than it needs to be. The risk, they warn, is that a real opportunity for AI safety could get buried under the politics of rivalry.
Source: Original report
Was this helpful?
Explore more: AI Automation Services More AI & Automation Tech News
Last Modified: September 24, 2026 at 10:32 pm
0 views
