China has pushed back on calls from U.S. technology leaders to slow the advance of frontier AI, while publicly insisting that rapid development and safety governance must proceed together. The issue gained a new pre-summit frame in a September 18 Nikkei Asia report, published days before the expected Trump-Xi meeting. The accessible Nikkei lead says that Anthropic CEO Dario Amodei’s warnings about existential risk had brought the AI rivalry into focus.
The key distinction is between resisting a particular capability-pacing proposal and rejecting AI safety work altogether. China’s Foreign Ministry has not said that safeguards are unnecessary. In an official September 15 response, spokesperson Guo Jiakun said China gives equal weight to development and security, takes AI risks seriously, and supports stronger governance to prevent systems from getting out of control. The official record supports a development-plus-security posture, not acceptance of an externally proposed slowdown.
Beijing’s response separates safety governance from capability pacing
Guo’s remarks came in response to a question about calls by U.S. technology leaders to slow AI development. He said that as AI advances quickly, it should be developed for positive purposes and for the benefit of people, while regulation and governance need to be calibrated to address risks. The Foreign Ministry also said that AI development and governance require multilateral cooperation, with the United Nations as the main channel for a global framework built on broad consensus.
Those statements are significant because they acknowledge risk while resisting the political framing around who should set the pace of frontier-model development. The official transcript does not say China accepted a global pause, a model-training moratorium, or a U.S.-led agreement. It does not announce that AI will be on a final leaders’ summit agenda. It does say that China wants continued development accompanied by security safeguards and multilateral governance.
Guo also referred to AI Safety Governance Framework 3.0, saying relevant authorities had released it the day before and that it updated AI-risk classification and included recommendations on technical countermeasures and comprehensive governance. The available official transcript does not establish the framework’s issuing authority, legal force, effective date, enforcement tools, or full contents. It is therefore safer to describe the framework as a named part of the Foreign Ministry’s presentation of China’s domestic governance approach, rather than as a new binding regulation.
EastFrontier has already covered China’s focus on loss-of-control risk in AI safety policy. The present development is not a repeat of that policy coverage. It is the emerging diplomatic friction between China’s stated governance model and a proposal to deliberately slow improvements in frontier-model capability.
Amodei’s proposal is an advocacy framework, not an international plan
Amodei’s own essay, “We Must Pace the Frontier,” defines pacing as taking sufficient time for alignment, safeguards, and third-party evaluation to keep up with advances in model capability. He says the proposal does not mean ending model training or halting technical progress. His three steps begin with a unilateral Anthropic commitment, move to industry coordination, and then contemplate global coordination.
That is an executive proposal from the head of Anthropic, not a U.S. government program or a negotiated international instrument. Its China dimension is also conditional. Amodei argues that global pacing requires cooperation with China, while warning that any agreement must preserve the technological lead of the United States and its allies. That combination helps explain why Beijing did not treat the proposal as a politically neutral safety initiative.
Reuters reported that Amodei had also urged stronger U.S. chip-export controls on China and action over alleged model distillation by Chinese labs. Those allegations should not be presented as established facts in a story about governance. They are relevant only as part of the political context surrounding the proposal and must remain attributed to Amodei or the reporting that described his position.
China’s official response and state-media commentary should also remain separate. Reuters reported Guo’s warning that fomenting threats, confrontation, and malicious competition would disrupt global AI governance. A Global Times opinion editorial went further, characterizing Amodei’s argument as a Cold War-style approach. The editorial is an opinion position, not a formal Foreign Ministry action. Combining the two would misstate the record.
The summit context remains uncertain, not a policy breakthrough
The exchange occurs against a longer sequence of attempted U.S.-China AI dialogue. Reuters previously reported, citing sources, that the two countries were planning to discuss frontier-AI safety risks in mid-September and that the subject could be raised at the Trump-Xi meeting. That remains reported and conditional. There is no verified final summit agenda, no joint statement, and no announced result.
This uncertainty is important because past coverage has shown how quickly safety language can collide with chip controls, model-access questions, and broader strategic competition. EastFrontier’s account of new calls testing the window for U.S.-China AI safety talks tracked an earlier attempt to identify common ground. The current dispute does not close that window, but it demonstrates that the parties attach different meanings to responsible governance.
China’s alternative framing is not simply domestic. At the September 13 BRICS summit, according to the Foreign Ministry’s English-language account, Xi Jinping said China would be a pioneer in forming a BRICS AI open-source community and support cooperation in developing and applying large language models. He also called for moving faster toward a consensus-based global AI-governance framework. That is an official statement of preferred international architecture, but it does not amount to a bilateral agreement with Washington.
For Washington, the debate around pacing raises questions about testing, safeguards, and the diffusion of powerful models. For Beijing, the same debate is entangled with access to chips, technology controls, and who gets to define global rules. The official Chinese response does not deny AI risk. Instead, it argues that development, security, and multilateral governance should be pursued together rather than through an externally imposed capability slowdown.
The immediate news is therefore not that China has rejected AI safety cooperation. It has not. Nor is it that a Trump-Xi meeting will produce guardrails. The evidence supports a narrower conclusion: a prominent U.S. company leader’s pacing proposal has sharpened the contrast between a strategy centered on deliberately slowing frontier capability gains and China’s stated preference for continuing development while strengthening domestic safeguards and UN-centered global governance.