a regular briefing
Dario Amodei’s pacing essay has three steps. The third one asks democratic governments to coordinate with authoritarian ones on how fast frontier models improve.
The answer arrived on Monday 15 September, at the Chinese Foreign Ministry’s scheduled press briefing, in response to a Reuters question. Spokesperson Guo Jiakun said that “fear-mongering, confrontation and vicious competition will only hamper efforts toward sound global AI governance.”
Note where that happened. Not a summit, not a white paper, not a counter-proposal with terms in it. A podium, on a Monday, on the regular schedule, to a reporter who raised it. The most consequential reply to the week’s AI-safety consensus was delivered as routine business.
That is not a rejection of the terms. It is a refusal to treat the premise as a subject.
Which leaves step three without a counterparty, and that matters more than it sounds, because a pacing agreement among a subset of participants is not a pacing agreement. It is a handicap. The whole logic of the essay is that unilateral restraint is bad — that a lab which slows down alone simply loses to one that does not — and the same argument applies one level up to a bloc.
Amodei is not naive about this. The essay calls for maintaining and strengthening export controls on advanced chips and chipmaking equipment, and for preventing model weight theft and distillation. He has been consistent on chips for years.
But read those together and the structure is plain. The coordination leg of this plan does not rest on diplomacy. It rests on hardware. The agreement part is for the democracies; for everyone else the mechanism is a supply chain you control and a security perimeter you can defend.
Which is a coherent strategy. It is just not what “global coordination” usually means, and calling it that invites exactly the reply Beijing gave — that the safety argument and the containment argument are the same argument wearing different clothes.
The hard part is that both things can be true at once. The risk can be real and the framing can also be self-serving. Guo’s line does not engage with the first, and the essay does not really engage with the second.
Sources: Dario Amodei, “We Must Pace the Frontier” ↗ · Bloomberg: China rejects AI “fearmongering” ↗ · TechNode Global ↗ · unite.ai on the MFA rebuke ↗