US proposes AI safety notification system to China before Trump-Xi summit

By Splitfeed EditorialLast updated Sep 26, 2026, 21:18 UTC

In talks held ahead of a planned Trump-Xi summit, Treasury Secretary Bessent formally proposed a mutual AI safety notification mechanism to China's vice premier around September 20 and 21, 2026. The proposal would create a channel through which each side could alert the other to AI-related national security incidents. Reporting indicates the two countries reached some form of agreement on an AI dialogue that includes a security component, though the arrangement as described excludes technical and scientific experts from its structure. China made no public statement acknowledging or endorsing the plan while the US side promoted it openly.

The context matters because US-China relations over AI have been defined largely by competition and mutual suspicion, making any formal communication channel significant in symbolic terms at minimum. At the same time, the substance of what was agreed is genuinely in dispute. Politico described internal skepticism about the hotline's practical value, with critics characterizing it as largely empty. The absence of technical experts from the mechanism raises questions about whether it could function meaningfully during an actual incident. Running alongside the diplomatic talks, a separate congressional debate over AI guardrails invoked China competitiveness explicitly, with the phrase "China's going to eat our lunch" circulating in that context, reflecting how domestic policy arguments and foreign policy moves are tangled together on this issue.

The Pro-AI camp has not reacted publicly to this story yet, but would typically be expected to argue that open AI development and American technological leadership matter more than diplomatic safety frameworks, and might view the proposal as bureaucratic friction on innovation. The Anti-AI camp has also not weighed in publicly, though a story like this would ordinarily prompt calls for more robust safeguards and concern that a notification-only mechanism is far too weak given the risks involved. Accounts in the Middle Ground camp, including worldnewsguru.bsky.social and American News Daily, highlighted the competitiveness framing with the "China's going to eat our lunch" language, suggesting the dominant concern in that space is strategic rivalry rather than the safety architecture itself.

The outcome of the Trump-Xi summit and any joint statement that follows would be the clearest next signal, either confirming the notification agreement in concrete terms or revealing how much of the reported deal holds up when leaders meet directly.