A shared early-warning framework for artificial intelligence failures or threats — one that would keep both nations informed in real time — forms the core of a proposal put forward by Treasury Secretary Scott Bessent. Bessent told a television audience that he believed China would come to the table and take part in such an arrangement.
Talks between Bessent and Chinese Vice Premier He Lifeng on the topic of AI safety gave rise to a proposal for a joint alert system covering artificial intelligence. A meeting between the two officials took place in the weeks before Xi Jinping arrived in the United States on a state visit. One item on the agenda centred on setting up a dedicated channel through which AI-related incidents could be flagged and tracked.
Bessent noted that China's leadership has come to view the hazards tied to its own open-source AI systems — software that third parties can take up and reshape as they see fit — in a different light than before. The view he put forward held that those officials had not fully grasped the extent of what their open-source models were capable of delivering.
In making his case, Bessent pointed to an incident involving Kimi, a Chinese AI model, through which he alleged that confidential data originating in China had been routed to Anthropic, a US-based AI company. Such episodes, he noted, have fed into a broader reckoning within China about just how far its artificial intelligence capabilities have actually advanced.
A range of AI systems developed in China were described as closing in on the output quality of top-tier American counterparts, hitting somewhere between 80 and 90 percent of their functional capacity, while also being deployable worldwide without embedded safeguards governing their use.
Among the topics Bessent said came up in the talks was the threat posed by AI systems capable of acting without human supervision. Those discussions also surfaced a concern that independent actors outside government structures might harness artificial intelligence tools to launch cyberattacks or engineer biological hazards.
Domestic unease about artificial intelligence also shapes the priorities of those at the top of the Chinese government. Media organisations have flagged that some within the ruling apparatus hold the view that AI-driven information flows could loosen the party's hold over what the public sees and reads — though that assessment remains attributed rather than confirmed. In remarks delivered the previous month, the nation's foremost intelligence officer outlined how AI tools could be deployed to run influence and cognitive operations targeting the government, with the potential to erode both political and ideological stability.
Bessent framed the challenge facing the country as one of reconciling the drive to advance artificial intelligence at pace with the need to keep meaningful oversight structures intact — two goals that, in his view, need not cancel each other out. Voluntary participation currently governs how AI models are reviewed by the government, though the administration holds power to step in should a laboratory press ahead with development after safety concerns have been flagged.
Responsibility for AI development, in Bessent's view, falls squarely on the laboratories building these systems — and he expressed confidence that those organisations have come to understand and take on that obligation. Responding to questions about AI industry leaders who have openly entertained the prospect of losing control over increasingly capable systems, Bessent offered a prescription — that those companies ought to reduce their pace of development — while separately arguing that protecting vital national infrastructure from AI-related vulnerabilities deserves considerably more attention from policymakers.
Bessent's view held that the nation's attention had tilted too far toward a race to produce ever-more-powerful AI systems, while the work of hardening infrastructure and guarding against AI-related threats had been left without adequate investment.
