Anthropic CEO Dario Amodei, OpenAI chief Sam Altman, and SpaceX AI founder Elon Musk have each publicly called for 'pacing' — a deliberate slowdown in frontier AI development — citing fears that capabilities are outrunning safety guardrails, a stance that would have been unthinkable even a year ago. The calls emerge ahead of a high-stakes summit between US President Donald Trump and Chinese President Xi Jinping in Washington on Thursday, 25 September 2026, raising urgent questions about whether geopolitical competition will make any voluntary slowdown viable.

Why It Matters

US Treasury Secretary Scott Bessent and Chinese Vice-Premier He Lifeng met in New York City over the weekend to discuss AI safety, signalling that the topic has moved squarely onto the diplomatic agenda between the world's two largest economies. The convergence of executive advocacy and presidential diplomacy marks a rare moment where the technology industry and government are publicly aligned on the need for restraint — at least rhetorically. Yet the structural incentives pulling in the opposite direction remain formidable. Both nations have framed AI supremacy as a core national interest, and neither has shown willingness to unilaterally cede ground in model capability, compute investment, or talent acquisition.

The Competitive Backdrop

China's response to the 'pacing' discourse has been closely watched. State media outlet People's Daily and affiliated commentators have historically framed US calls for AI governance as thinly veiled attempts to lock in American technological advantages — a reading that complicates any genuine bilateral coordination. For Beijing, agreeing to slow down risks conceding hard-won ground in a race it believes it is narrowing. The phrase 'constructive strategic stability' — the framing both governments have agreed to work toward — offers diplomatic cover but little operational clarity. What counts as a 'safe' pace of development, and who arbitrates that standard, remains entirely unresolved.

The Pacing Argument Unpacked

The executives' shared concern centres on a specific claim: that frontier AI systems are acquiring capabilities faster than the interpretability tools, alignment techniques, and regulatory frameworks needed to govern them. Amodei, Altman, and Musk have each articulated versions of this argument in recent weeks, lending it unusual cross-industry weight given their otherwise competing commercial interests. Critics, however, note that 'pacing' as articulated by incumbents could function as a moat — slowing challengers, domestic and foreign, while entrenching the position of labs already at the frontier. The dual-use nature of the argument makes it difficult to separate genuine safety concern from competitive strategy.

What's Next

All eyes are on the Trump–Xi summit in Washington on Thursday to see whether the two leaders can translate the Bessent–He Lifeng talks into a concrete AI safety framework — or whether the meeting produces only broad communiqués with no enforcement mechanism. The outcome will determine whether 'pacing' remains an industry talking point or becomes the foundation of the first bilateral AI governance agreement between the United States and China. The most exposed stakeholders are frontier AI labs operating in both markets, defence-adjacent AI contractors, and semiconductor suppliers whose order books depend on the pace of model scaling continuing unabated.