• The U.S. has floated a plan to set up a direct notification channel with China for severe AI-related incidents that threaten national security, Treasury Secretary Scott Bessent said.
  • The proposal, discussed with Chinese Vice Premier He Lifeng in New York, is in early stages and would be part of a broader U.S.-China AI Dialogue.
  • Advanced-chip export controls and semiconductor equipment restrictions remain outside the scope, signaling limited cooperation amid ongoing strategic rivalry.

A Narrow Opening

Treasury Secretary Scott Bessent has proposed establishing a dedicated communication line with China to manage AI-related incidents that could escalate into national security crises, according to people familiar with the matter. The idea, raised during talks with Chinese Vice Premier He Lifeng in New York, aims to improve transparency between the world’s two leading AI powers and prevent misunderstandings that might spiral out of control.

The proposed mechanism would cover AI incidents that “rise to a national security level,” rather than routine technical failures or commercial disputes. Bessent framed it as a way to identify “common goals and common threats” and to keep the two countries’ AI systems from bifurcating further.

China has acknowledged that AI was discussed but has not publicly accepted the proposal, disclosed a negotiating position, or agreed to any operating rules, definitions, or timelines. In practical terms, this is an early-stage confidence-building measure—not an operational hotline yet.

Carve-Outs and Caveats

Notably, the talks explicitly excluded advanced-chip export controls and semiconductor-manufacturing equipment restrictions. That separation signals Washington wants limited safety cooperation while maintaining strategic technology restrictions.

The discussions did touch on both open-weight AI models—whose underlying components are publicly available—and closed-weight models. But the proposal’s scope remains narrow: it would not address broader trade tensions, tariffs, or export controls.

A U.S. official said the plan is still under development and could change. The White House and Treasury declined to comment on the record. Chinese officials did not respond to requests for comment.

Pressure from All Sides

The initiative comes amid a broader push to stabilize economic ties between the world’s two largest economies. The New York talks also covered tariffs, trade, rare earths, and critical minerals. The existing U.S.-China tariff truce is due to expire on November 10, increasing pressure on both governments to find areas of limited cooperation.

China’s leverage in rare-earth materials and magnets has become central to negotiations. These inputs are critical for autos, aerospace, high-tech manufacturing, and defense supply chains; disruptions could raise costs and delay production outside China.

Washington wants more reliable mineral supplies, while Beijing has sought a further delay to a U.S. rule that would restrict thousands of Chinese firms’ access to advanced U.S. technology.

Market Implications

For investors, the clearest near-term implication is modestly constructive: a dedicated AI-risk channel could reduce the tail risk of abrupt policy escalation or misinterpretation after a major AI incident. It does not, however, indicate a rollback of U.S. controls on advanced AI chips or manufacturing tools.

“Any mechanism that reduces the chance of a miscalculation is welcome, but it’s no substitute for a broader framework,” said one person familiar with the discussions. “This is about guardrails, not détente.”

U.S. frontier-AI developers could eventually face expectations to report severe cross-border AI security events, while Chinese AI developers may encounter parallel reporting demands if Beijing accepts a mechanism. Chipmakers and equipment firms remain mainly affected by export-control policy, which was kept outside this proposal.

Political Timing

The timing is highly political. Bessent and He’s talks were designed to prepare the ground for a planned September 24 meeting between Presidents Donald Trump (DJT) and Xi Jinping, where both leaders may seek incremental, deliverable outcomes rather than a broad settlement of U.S.-China technological competition.

Bessent has emphasized a balance between innovation and safety, while indicating that the administration does not currently see a specific new AI law as the primary answer; the focus appears to be executive-branch engagement with labs and financial-sector partners.

The proposal is akin in purpose to crisis-communication arrangements in other high-risk domains: prevent an incident from being mistaken for deliberate escalation, create a way to verify facts, and give officials a channel before political or military reactions harden.

Yet this is not a détente in the technology contest. Export controls, access to high-end compute, chip supply chains, Taiwan-related tensions, trade tariffs, and critical minerals remain separate disputes.

What to Watch

The principal test is whether Trump and Xi endorse further work on the proposal at their September 24 summit. A likely initial outcome, if any, would be a commitment to continue dialogue—not a fully defined hotline, common incident taxonomy, or enforceable reporting regime.

Investors and companies should watch for language defining “national-security-level” AI incidents, naming responsible agencies, and setting a timetable for technical meetings. Those details would determine whether the plan becomes operational.

In the best case, a standing channel reduces the likelihood that an AI-enabled cyberattack, model-security failure, infrastructure disruption, or false attribution event triggers a U.S.-China confrontation. A more likely scenario is that the dialogue becomes a narrow crisis-management forum alongside ongoing rivalry over chips, models, cloud capacity, military applications, and supply chains. The downside case: disagreement over verification, secrecy, attribution, or what constitutes an “AI incident” prevents implementation, leaving the proposal as summit diplomacy rather than a real safety mechanism.

The key analytical point is that transparency on extreme AI risks and continued competition for AI advantage can coexist. The proposal does not suggest that Washington and Beijing are converging on AI governance overall; it suggests they may be testing whether even strategic competitors can establish a minimal guardrail against accidental escalation.

Correction: An earlier version of this article misstated the date of the planned Trump-Xi meeting. It is scheduled for September 24.