US and China open AI safety talks as rivalry deepens

Treasury Secretary Scott Bessent said Washington and Beijing will begin discussing artificial-intelligence safety, opening a formal channel between the two governments even as they remain locked in a wider contest over chips, security and global tech influence.

Both sides see enough shared risk in frontier AI systems to talk guardrails, though neither appears ready to ease the broader strategic competition that has defined the relationship. The move is an attempt to put limits around one part of a rivalry that is still expanding — not a thaw. AI safety has emerged as a subject where both capitals can discuss common danger without pretending the commercial or strategic fight has been settled.

Speaking after the Trump-Xi meeting, Bessent said the two countries would discuss a protocol to keep the most powerful models out of the hands of nonstate actors, the New York Times reported. “The two A.I. superpowers are going to start talking,” Bessent said, according to the report. In separate remarks to CNBC, he paired that opening with a blunt assessment of Washington’s position: “We are in the lead.”

If the talks proceed, they would be the first formal U.S.-China AI discussions of Donald Trump’s second term, the Times reported. The two-day Trump-Xi summit had already become a venue for managing disputes over Taiwan, rare earths, Nvidia chip sales and the wider effort to keep the relationship from sliding from confrontation into rupture. AI safety now joins that list of tightly managed subjects where competition and stabilisation can coexist.

Reuters reported that the proposed talks would centre on guardrails for the most advanced systems, not on reopening the harder fights over export controls or market access. Bessent’s CNBC remarks drew the same line. Officials can discuss shared downside risk while leaving the contest over computing power, chip tools and military-adjacent applications intact. The focus on frontier models is narrower than the broader dispute over AI competition, pointing to the systems whose misuse carries the highest stakes rather than the daily policy fights around consumer applications or routine industrial deployment.

Controlled competition

The framing puts AI closer to a strategic-stability issue than a consumer-technology story. Both governments increasingly treat these tools as national-security assets, not consumer products. What matters is how both capitals handle systems they see as potentially dangerous — not which chatbot leads the monthly download charts.

A safety channel does not mean the technology contest is easing. The Washington Post described the moment as part of a broader AI cold war in which the United States and China are trying to shape standards, secure supply chains and protect their own lead in key systems. Digitimes reported that recent exchanges between the two sides have mixed selective co-operation with sustained pressure over semiconductors, export controls and Taiwan-related security concerns.

The channel sits alongside the rest of the agenda, not in place of it. Both governments get a way to discuss one category of shared risk while leaving intact the competition over advanced chips, strategic industries and the military implications of faster AI development. For technology companies and investors, the distinction is consequential: a government-to-government conversation on AI safety may lower the odds of an uncontrolled policy shock if both sides start setting expectations around frontier models. It does little, however, to change the immediate commercial landscape, where Chinese access to top-end chips and U.S. concerns about dual-use technology remain central.

Why would both capitals want a channel this narrow? Washington can argue it is reducing the risk of the most dangerous AI misuse without relaxing pressure on advanced chips or giving up its edge. Beijing can enter talks that recognise its status as a leading AI power without winning broader concessions on the controls and restrictions that still dominate the commercial side of the relationship. Much remains unclear — when the talks will begin, who will lead them, whether they will produce anything more binding than an exchange of views. The significance for now is the creation of a channel, not any finished agreement. Even at a moment of deep distrust, both governments see at least one area where rivalry requires rules as well as pressure.