Treasury Secretary Scott Bessent wants Washington and Beijing to establish a way to alert each other when artificial intelligence goes wrong, opening a potential channel for cooperation between two countries competing to dominate the technology. Bessent said he believes China would agree to an emergency notification process.

Bessent discussed AI safety with Chinese Vice Premier He Lifeng before President Xi Jinping’s state visit last month. Their conversations included uncontrolled AI agents and the possibility that groups outside government could use the technology to launch cyberattacks or create biological threats. Bessent said Beijing has begun recognizing the dangers posed by its powerful open-source models, which users worldwide can modify.

He cited an incident in which China’s Kimi model allegedly sent sensitive Chinese information to Anthropic, saying it helped expose risks Chinese officials had underestimated. Bessent is not alone in calling for more regulation of AI, as the industry’s biggest leaders have repeatedly called for more regulation as they develop increasingly capable systems.

Bessent urged developers to slow down if they fear losing control, while calling for greater protection of critical systems as America races to build ever more powerful AI models.

The thought of an AI crisis or emergency that spreads across borders before governments understand what is happening is hardly far-fetched. These systems have access to sensitive data and have the capability to disrupt infrastructure, and even develop a biological weapon. Such a scenario gives Washington and Beijing a strong incentive to work towards a solution together, regardless of their rivalry.

A notification channel could help officials identify a problem sooner and coordinate a response before damage spreads. Bessent is pushing for the United States and China to establish an emergency notification channel so officials can alert each other when AI systems go wrong, reflecting strong concerns about powerful models that anyone can download and modify, potentially weakening safeguards and putting dangerous capabilities within reach of criminals or other hostile actors.

But for this channel to be useful, both governments would need to agree on which incidents require an alert and share information quickly enough to contain the damage. While Bessent believes Beijing would participate, neither country has committed to an arrangement yet, leaving it unresolved how much either would disclose during a crisis.

Bessent’s warning is likely to tangibly affect how AI moves through the workforce. If developers heed his call to slow down when control becomes uncertain, employers that increasingly rely on capable systems may have to reconsider how soon they can automate tasks or reduce staffing. Although that would not guarantee anyone’s job, it could complicate plans built around technology improving on schedule.

Bessent’s comments suggest Washington could step in even when an AI company is willing to keep developing a system. For businesses adopting AI, this creates uncertainty regarding how quickly new tools arrive and how much work companies can hand over to them. If the Trump administration follows through with Bessent’s proposal, these warnings can become grounds for intervention.