The most revealing part of the latest Donald Trump-Xi Jinping encounter may not be what the two leaders agreed on. It may be what they could not agree on: how to manage a technology that neither Washington nor Beijing is willing to slow down.
Artificial intelligence has become the newest arena of great-power competition. The United States wants to preserve its technological lead. China wants to close the gap and shape the rules governing a technology it increasingly sees as central to its economic and strategic power. Between those objectives lies a problem that cannot be solved by competition alone.
AI does not recognise national borders.
A system capable of conducting sophisticated cyber operations, manipulating information, controlling critical infrastructure or assisting military decision-making can create consequences far beyond the country in which it was developed. That makes AI governance fundamentally different from a conventional technology race.
The Trump-Xi summit demonstrated how difficult that reality is becoming.
Both leaders have acknowledged the need to discuss AI. China has pushed the principle that AI must remain under human control, while Washington has placed greater emphasis on maintaining technological leadership and avoiding restrictions that could slow American companies.
Those positions are not necessarily irreconcilable. But they reveal the central contradiction of the AI age: the same governments that are increasingly concerned about the technology's risks also believe that falling behind in its development could create an even greater strategic danger.
For Washington, the fear is straightforward. If the United States imposes constraints that China does not follow, American companies could lose their technological advantage. For Beijing, allowing the United States to dominate advanced AI would carry its own economic and strategic consequences.
The result is an arms race in which safety can easily become another dimension of competition.
That is where the world should be worried.
The danger does not require an artificial general intelligence breakthrough or a science-fiction scenario involving machines taking control of governments. Much more immediate risks are already visible: autonomous cyber activity, unreliable AI-generated intelligence, manipulation of information, attacks on critical infrastructure and increasingly autonomous systems operating in military environments.
The more capable these systems become, the harder it may be to determine whether an unexpected action was deliberate, accidental or generated autonomously.
That ambiguity is particularly dangerous between two strategic competitors.
During the Cold War, Washington and Moscow eventually recognised that communication mechanisms were necessary even when trust was almost nonexistent. Hotlines did not end the rivalry. They provided a way to prevent misunderstandings from becoming catastrophes.
AI needs an equivalent logic.
The United States and China do not have to trust each other to establish basic rules for dealing with an AI-related crisis. They do not need to agree on every aspect of regulation, surrender technological advantages or stop competing for commercial leadership.
They do need mechanisms that allow them to communicate when an advanced AI system behaves unexpectedly or when one side suspects the other of using AI in a way that could trigger escalation.
That is a much narrower objective than creating a universal AI treaty. It is also potentially more realistic.
The problem is that even modest cooperation is becoming entangled with the broader technology rivalry.
Washington's restrictions on advanced semiconductor exports to China are designed to constrain Beijing's access to some of the computing resources needed for frontier AI. Beijing, meanwhile, is building its own ecosystem and seeking greater influence over international AI standards.
Both sides therefore have reasons to regard technological openness with suspicion.
Yet the assumption that competition and safety are mutually exclusive may prove increasingly dangerous.
There is a difference between cooperating with a rival to prevent catastrophic misuse and cooperating with that rival to strengthen its technological capabilities. Governments can protect strategic technologies while still agreeing on minimum safety standards, crisis communications and procedures for investigating AI-related incidents.
That distinction will become increasingly important.
The AI race is also forcing countries outside the US-China rivalry to confront an uncomfortable question: who gets to write the rules?
China has been promoting its own vision of international AI governance and building institutions through which it can expand that influence. The United States has historically had enormous technological and institutional influence, but its approach to AI governance has been less unified.
If Washington and Beijing turn AI governance entirely into another contest for geopolitical influence, the rest of the world may be left choosing between competing systems rather than helping shape common standards.
That would be a mistake.
The countries developing the most advanced AI systems have an obvious responsibility to manage the risks they create. But countries that are not leading the technological race will also bear the consequences of failures.
India, Europe, Japan, South Korea and other major technology users therefore have a stake in ensuring that AI governance does not become a closed conversation between Washington and Beijing.
The immediate objective should not be to end the US-China AI competition. That is neither realistic nor necessarily desirable.
It should be to prevent competition from eliminating the guardrails needed to keep that competition from becoming dangerous.
The irony is that the United States and China may eventually discover that AI is simultaneously a weapon of strategic competition and a reason for strategic restraint.
Neither side needs to surrender its ambitions.
But both need to recognise that some risks cannot be defeated by being faster than the other side.
If an AI system launches a cyberattack, generates false military intelligence or triggers an automated response, the question of whether the machine was American or Chinese may become irrelevant.
The consequences will be global.
The real test of the Trump-Xi AI dialogue, therefore, will not be whether Washington and Beijing can agree on who should lead the technology.
It will be whether two competing powers can agree on one basic principle: winning the AI race means little if neither side can control where the race ends.
