They met in New York

They met in New York. On Sunday, America’s Scott Bessent and China’s He Lifeng began high level discussions on a topic of existential weight. The subject was artificial intelligence. Specifically, it was about how the world’s two greatest powers might prevent an AI from triggering a catastrophe. This was not a routine dialogue on trade deficits or intellectual property, topics also on the agenda. It was a carefully choreographed engagement designed to build a framework for managing a technology that could spiral out of control with terrifying speed, a recognition from both Washington and Beijing that their fierce competition requires new rules of the road.

The timing was deliberate. This was a prelude. The discussions in New York are intended to clear a path for a formal agreement when President Trump meets President Xi later this week. Both sides want a result. They seek to establish a dedicated communication mechanism, a hotline for the digital age, to manage potentially serious AI incidents before they escalate. It is a sign that, even as the two countries wrestle for economic and military supremacy, a shared sense of vulnerability has entered their calculations. The rivalry has not been set aside. It has simply acquired a new, more complex dimension.

Washington and Beijing are acutely aware of the dangers. The talks were a direct result of this awareness. An AI that precipitates a stock market crash, a military system that misinterprets data and advises a pre-emptive strike, or a state backed disinformation campaign that escapes the control of its creators are no longer the stuff of fiction. They are plausible strategic shocks. The presence of Bessent and He Lifeng, senior figures entrusted with immense economic and technological portfolios, signals the gravity with which both governments now treat these risks. This meeting was not for junior officials. It was for principals. It was for building guardrails around a precipice. The stakes are immense.

This is about managing competition

This is not friendship. It is risk management. The talks in New York do not signal a thaw in the icy relationship between Washington and Beijing, nor do they represent a pause in the relentless competition for technological dominance that has defined geopolitics for the past decade. This is not peace. The opposite is true. The discussions are a direct consequence of the hostility, an admission by both sides that their rivalry has entered a phase so unpredictable that it requires new and specific safety measures to prevent a global catastrophe. The race for AI supremacy continues unabated, with billions of pounds channelled into research and development from Shenzhen to Silicon Valley, but now it will happen with guardrails. This is an attempt to build a fence at the top of the cliff, not an ambulance service at the bottom. The fundamental conflict over semiconductors, trade routes, intellectual property and military influence has not been resolved or set aside. It has simply become so dangerous that its primary architects now seek to insure themselves against its worst possible outcomes.

The historical parallel is precise. It is from the Cold War. The discussions between Scott Bessent and He Lifeng echo the logic that followed the Cuban Missile Crisis in October 1962, a moment when the United States and the Soviet Union stared into the abyss of mutual nuclear destruction. Out of that terror came a pragmatic realisation. It led to the establishment of the Moscow Washington hotline in 1963, a direct teletype link designed to give the leaders a chance to clarify intentions and de-escalate a crisis before missiles were launched on the basis of faulty information. That system was not built on trust. It was built on a shared fear of extinction. The proposed AI communication mechanism is a digital successor to that teletype machine, a recognition that a new technology, one that operates at the speed of light and can learn and adapt on its own, presents a similar risk of accidental, world ending escalation.

Artificial intelligence presents a unique challenge to this old model of strategic stability. A nuclear arsenal is a physical thing. Its warheads can be counted by spies and satellites, and its launch requires a series of deliberate, observable human actions. An AI is different. An AI is code. A rogue algorithm causing a financial flash crash or a military drone misidentifying a target could trigger a chain reaction that unfolds in milliseconds, far too fast for human leaders to convene, consult and respond through a traditional diplomatic process. The source of the disruption might not even be immediately clear. Was it a deliberate attack, a catastrophic system error or something in between? The ambiguity is itself a profound source of danger. The New York talks are an attempt to create a vocabulary and a channel for precisely these scenarios. They are about creating a pause.

Ultimately, the dialogue is an act of supreme self interest for both powers. China wants to become the world’s AI leader by 2030. America is determined to stop that from happening. Nothing about the meeting in New York changes these fundamental strategic objectives. Neither side is offering to limit its research or share its breakthroughs. The contest will go on. The goal is simply to ensure the contest itself does not burn down the stadium with all the spectators inside. The creation of a communication channel is an insurance policy against the potent and unpredictable power they are both racing to unleash. It is an agreement to talk, nothing more. Survival is the only shared goal.

The men in the room

The men sent to New York were not diplomats. They were not generals. Their presence signalled that the immediate anxieties in Washington and Beijing are not primarily military, but economic. The choice of negotiators reveals everything. Both sides picked their money men.

America sent Scott Bessent. He is a creature of Wall Street, a former hedge fund manager from the school of George Soros, and the man widely expected to become Donald Trump’s next treasury secretary. His entire professional life has been about quantifying uncertainty and pricing risk, a skillset honed managing billions of pounds in volatile global markets where a misplaced algorithm could wipe out a fortune in seconds. Bessent’s appointment frames the AI safety problem in terms Washington’s elite understands. The fear is a market crash. The nightmare is a financial contagion sparked by autonomous trading systems, an event that moves too fast for any human intervention. By sending its chief economic risk manager, the White House is showing that its most pressing concern is preventing a catastrophic failure within the world’s deeply interconnected financial architecture. The talk is of guardrails, not missile shields.

His opposite number was He Lifeng. He is a different kind of economist, a powerful Communist Party official who sits on the 24 member Politburo and holds the title of vice premier. He is the man in charge of China’s sprawling economic and financial system, an architect of the state directed industrial policy that has identified artificial intelligence as a critical engine for future growth. He Lifeng is President Xi Jinping’s trusted planner, the bureaucrat tasked with ensuring China achieves its goal of global AI leadership by 2030. His attendance demonstrates that for Beijing, AI is a central pillar of national strategy and economic security, inseparable from its ambition to become the world’s dominant power. It is about control. It is about development. The discussion is not a technical sideshow for China, but a core component of its grand economic vision, requiring oversight from the very top of the party state. This is about shaping the future.

What counts as an 'AI incident'?

What counts as a serious AI incident? The term is deliberately broad. It has to be. The threats are new. Some are still theoretical, scenarios sketched out in policy papers and war games in Washington and Beijing. The shared fear is speed. An AI crisis will not unfold over days like the run on Northern Rock, or even hours like a stock market flash crash, but in the milliseconds that leave no time for a considered human telephone call. A direct line between the two powers would need to address events that escalate much faster than any person can track, process, or understand. It is about preventing catastrophe.

The most immediate danger is financial. Consider the global markets where trillions of pounds are managed by high frequency trading algorithms. These automated systems react instantly to data feeds, news headlines, and the movements of other algorithms across exchanges from Shanghai to London. A hostile state or even a non state actor could introduce a false but convincing signal, maybe a hyper realistic deepfake video of a central banker announcing an emergency interest rate rise, triggering a cascade of automated selling that bottoms out a currency. The contagion would be global. It would be instant. A pre agreed communication channel is meant to be a brake, a way for one side to tell the other ‘our systems are seeing an anomaly, it is not real, do not react’.

The military risks are more direct. They are more terrifying. Both the United States and China are pouring billions into developing AI enabled weapons and command systems, from autonomous drone swarms to software that predicts enemy movements. Imagine an AI powered early warning system guarding the Taiwan Strait misidentifying a civilian airliner’s electronic glitch as the signature of an incoming stealth bomber attack. The system, designed and authorised to react at machine speed to shorten the chain of command, could launch a retaliatory strike without final human approval. An immediate, trusted explanation via a secure channel might be the only thing that stops a terrible regional mistake from spiralling into a full scale war between two nuclear armed powers. This is an old Cold War problem. It is now on hyperspeed.

Then there is the political realm, where AI driven disinformation threatens to shatter social cohesion. A hostile AI could be tasked with destabilising a rival nation’s democracy during a tense election period. It could generate thousands of unique, believable videos appearing to show ballot tampering and distribute them through armies of automated social media accounts, precisely targeting key demographics to incite rage and distrust. Such an operation could provoke real world violence. It could paralyse a government. The goal is managing chaos. A dedicated channel would allow one side to alert the other that a massive, state level information attack appears to be underway, creating a sliver of time to verify its origin before a society tears itself apart.

A fragile first step

The meeting in New York is a prelude. The main event is the summit between Donald Trump and Xi Jinping later this week. What Scott Bessent and He Lifeng discussed on Sunday will be packaged for their presidents, a tangible outcome from a relationship defined by friction. It is a sign of progress. It is a small one. The creation of a communication mechanism for AI incidents, if it is formalised, represents the bare minimum of risk reduction in a field defined by its boundless potential for chaos. The model is the Moscow Washington hotline, a Cold War innovation designed to stop two superpowers from stumbling into nuclear annihilation through misunderstanding, accident or technical error. This is a digital echo of that logic.

This new version will not be a red telephone in a bunker. It will almost certainly be a secure, encrypted digital channel connecting military and civilian crisis centres. But its function is identical. Its effectiveness, however, is far from guaranteed. A hotline works only if both sides believe what they are being told, a monumental challenge when the events being described are generated by opaque algorithms operating at speeds beyond human comprehension. How can Beijing prove to Washington that a sudden, ruinous flash crash on the New York Stock Exchange was caused by its own rogue trading AI and not a deliberate act of economic warfare? How can trust be established? It cannot be coded. The problem of verification, of distinguishing an accident from an attack, may be technically impossible.

The obstacles are immense. The very definition of a crisis is itself a point of contention between Washington and Beijing. For the United States, the primary fear is a military miscalculation in the Indo Pacific, an AI error in the Taiwan Strait that leads to a shooting war. For China, the greater threat might be an AI used to fuel dissent or separatism within its borders, something it would view as a grave attack on its sovereignty and social stability. Agreeing on what specific events warrant an emergency call will be a fraught and lengthy negotiation. It will be long. The fundamental rivalry has not been paused. This meeting does not end the technology war.

This is a narrow path. It is a specific solution for a specific problem. Washington will continue to restrict China's access to high end semiconductors, the silicon brains that power advanced AI. Beijing will continue its massive state led effort to achieve technological self sufficiency and overtake the United States. They are still competitors. They are still adversaries. The race for AI supremacy, which both nations see as the key to future global power and economic prosperity, will accelerate, not slow. A communication channel is simply an admission of shared vulnerability, a joint recognition that the tools they are building to win this contest could, if they malfunction, destroy the prize for everyone. It is a fragile first step. It is not the last.

Sources. Al Jazeera: US, China open high-level talks ahead of Trump-Xi summit. France 24: US, China discuss AI safety ahead of Trump-Xi summit.

Analysis. Drafted with AI assistance from the sources listed above and reviewed by an editor before publication. Jnews links to the organisations it writes about.