The US and China Are Building AI Guardrails While Competing Everywhere Else

The most consequential part of the latest US-China diplomacy may not be a tariff concession or a purchase agreement. Washington and Beijing are exploring how they would communicate when an artificial-intelligence incident becomes serious enough to create national-security risk. That is a modest mechanism by the standards of superpower diplomacy. It is also evidence that the AI rivalry is entering a more mature and more dangerous phase.

Great powers build guardrails not when competition disappears, but when competition becomes important enough that miscalculation carries unacceptable costs. The United States and China remain divided over advanced semiconductors, export controls, rare earths, Taiwan, trade and the distribution of technological power. The emergence of an AI risk channel should be read against that rivalry, not as an alternative to it.

Competition is creating the need for communication

The logic is familiar even if the technology is new. Rivals that distrust one another can still share an interest in preventing accidents, false alarms or uncontrolled escalation. During the Cold War, Washington and Moscow developed hotlines, arms-control verification and incident-management mechanisms without resolving their underlying ideological and military competition.

AI is not a nuclear weapon and the analogy should not be stretched. The relevant similarity is institutional: both governments are confronting a technology whose failures can cross borders faster than ordinary diplomatic processes. A serious cyber event involving an autonomous system, a model used in military decision-making or a rapidly propagating AI-enabled attack could leave leaders trying to distinguish accident, private action and deliberate state behaviour under time pressure. Communication has value before trust exists.

The guardrail itself will become part of the power struggle

There is a harder problem. The two countries do not agree on what responsible AI governance should protect. Washington's export controls are intended in part to restrict Chinese access to frontier computing capability. Beijing regards important elements of American technology policy as containment. Rules about model evaluations, incident thresholds or information sharing can therefore become arguments about sovereignty and industrial advantage.

State & Power's view is that this makes a narrow agreement more plausible than a grand AI treaty. An incident-notification channel can define a small category of events that both governments would rather understand than exploit. Attempts to harmonise the wider AI order would immediately collide with surveillance, censorship, military use, export controls and competing concepts of state authority.

Why this matters beyond Washington and Beijing

A bilateral mechanism between the two largest powers would inevitably influence other states. Europe, Britain, Japan, South Korea and major AI-producing economies are building their own safety and security regimes. If the United States and China establish even a limited vocabulary for severe AI incidents, other governments will face pressure to decide whether to adopt, modify or reject it.

That creates an unusual form of institutional power. The state that helps define what counts as a reportable AI incident can shape expectations far beyond its own borders. Standards that begin as crisis management can eventually influence procurement, model testing, cyber policy and corporate disclosure.

The real test is whether the mechanism survives a crisis

Diplomatic announcements are easy when leaders are in the same room. Useful guardrails are built in the months afterward: named points of contact, agreed thresholds, exercises, technical communication and procedures that continue operating when the political relationship deteriorates.

That is the metric we would use here. The important question is not whether Trump and Xi can describe AI safety as a shared interest. It is whether their governments create a channel officials would actually use during the kind of ambiguous, politically dangerous incident that makes communication hardest. If they do, strategic rivalry will not have weakened. It will have acquired rules for surviving itself.