US Touts AI Safety Alert Channel With China as Trust Runs Thin

The United States and China, the two global leaders in artificial intelligence, opened talks this week intended to protect the world from emerging AI safety risks — and Treasury Secretary Scott Bessent says the two sides have already discussed a concrete first step: a new AI safety notification mechanism.

Bessent announced the proposal ahead of President Donald Trump's meeting with Chinese President Xi Jinping on Thursday and Friday. If China agrees to participate, the mechanism would create an open line for either country to send alerts when its AI systems pose threats or behave unpredictably, according to Bessent. He said the same channel could allow the countries to share their "vision of common goals."

That is the stated ambition. Whether the mechanism becomes anything more than a diplomatic gesture depends, according to experts cited in Ars Technica's reporting, on how much two fierce rivals are genuinely willing to cooperate — and on whether meetings between Trump and Xi can produce a reliable global governance framework for AI at all.

A channel that leaves experts on the outside

One detail of the US proposal stands out immediately: as Ars Technica reported, the plan for AI safety alerts omits tech experts. China, for its part, has remained silent on the proposal. The US is touting the plan; Beijing has not signed on in public.

That silence matters. A notification mechanism only functions if both parties trust that the information flowing through it will not be weaponized. An alert about a model behaving unpredictably is, in effect, a disclosure about the state of a nation's most sensitive technology program. Experts quoted in the report warn that framing AI policy as a race — a framing the White House has embraced — may itself endanger the United States, because it treats safety cooperation as secondary to competitive advantage.

The story's core tension is straightforward. The US wants to be seen leading on AI safety coordination with its chief strategic competitor. At the same time, US policy is actively designed to constrain that competitor's AI capabilities. Both things can be true at once, but they sit uneasily together at a negotiating table.

Why the timing is difficult

China may not be ready to trust the US to hold AI safety conversations that are not, at their core, about limiting China's capabilities. The reasons are concrete and recent.

First, Trump has steadily increased export controls that explicitly seek to block China from accessing technology to advance AI. The restrictions are not incidental to the relationship; they are a stated instrument of US strategy.

Second, Congress is considering bills that would cut China off from additional chips and hardware. Lawmakers are debating measures that would extend the existing restrictions deeper into the semiconductor supply chain, according to the report.

Third — and most inflammatory — just two weeks before the Trump-Xi meeting, Trump's Justice Department accused six of China's leading AI makers of stealing from US frontier labs. China has strongly disputed the accusations. The case, covered by Ars Technica, names some of the most prominent firms in China's AI sector and alleges they aggressively copied US frontier models.

Put together, the sequence looks like this from Beijing's perspective: Washington tightens technology restrictions, threatens further cutoffs in Congress, and files criminal accusations against six leading Chinese AI companies — then proposes an AI safety hotline and asks for trust. That is the diplomatic terrain on which any notification mechanism would have to be built.

Why this matters beyond Washington and Beijing

The stakes are global. The US and China are the two world leaders in AI, and the systems they are developing carry risks that do not respect borders. A model that behaves unpredictably in one country's lab can produce consequences — malicious use, loss of control, cascading failures — that spill into everyone's economy and information environment. That is why the talks this week were framed from the outset as an effort to keep "the whole world safe" from emerging AI risks.

A functioning notification mechanism between the two leading AI powers would be, at minimum, a precedent. No comparable permanent channel for government-to-government AI incident alerts currently exists between rivals of this scale, the reporting indicates. Even a modest, narrowly scoped line of communication — limited to alerts about dangerous or erratic model behavior — would establish that safety information can move between strategic competitors without being reduced to leverage.

The alternative, in the view of the experts cited, is that the "AI race" framing wins. If both governments treat every AI interaction as zero-sum, safety information stays bottled up, and the world gets the risks of frontier AI without the coordination needed to manage them. The experts' warning is blunt: Trump's China rivalry and what they call the "AI race" delusion may endanger the US itself.

The optics problem

Trump clearly wants the discussions to appear productive. The Bessent announcement came before the summit, not after — an unusual sequencing that signals the administration prioritized demonstrating momentum. Announcing that both sides had "already discussed" the mechanism, before Trump and Xi had even met, served that purpose.

But announcements are not agreements. China has not confirmed participation. Bessent's description of the channel — alerts when systems "pose threats or behave unpredictably," plus a venue for sharing a "vision of common goals" — remains a US characterization of a proposal, not a joint commitment. The gap between how the US is presenting the mechanism and how China is receiving it is, at this stage, the story.

The omission of technical experts from the proposed structure compounds the credibility question. AI safety alerts are only useful if the people sending and receiving them can assess technical detail — the nature of a failure, the scale of a threat, the reliability of a warning. A purely government-to-government channel staffed by officials without technical depth risks becoming a diplomatic formality, critics of the approach imply in the reporting.

What to watch

The immediate indicator is whether China publicly joins the mechanism in the aftermath of the Trump-Xi meeting. Silence, which is Beijing's current posture, would leave the US touting a one-sided proposal — precisely the situation the pre-summit announcement was designed to avoid.

The medium-term indicator is whether the Justice Department accusations against the six Chinese AI firms and the pending chip legislation in Congress harden Beijing's refusal to engage. Any of those threads — export controls, congressional bills, the criminal case — could give China reason to treat the safety channel as a trap rather than a tool.

The longer-term question is the one the experts pose directly: can a reliable global governance framework for AI emerge from a relationship defined by rivalry? This week's talks opened the possibility. Whether Trump and Xi convert it into anything durable depends, as the reporting concludes, on how much the two rivals are actually willing to cooperate — and every action Washington has taken in the weeks before the summit has made that calculation harder for Beijing, not easier.