Can the US and China “verify” AI restraint—or will supremacy rules break the deal?

On September 23, 2026, multiple outlets converged on a single dilemma: Washington and Beijing are being urged to slow AI development, yet the core question is verification—how can either side prove the other is actually complying. A piece from SCMP highlights researchers and start-ups building tools aimed at making AI-deal verification feasible, implying that “monitoring” could become a new diplomatic instrument rather than a purely technical afterthought. BBC framed the same day’s US–China engagement as a contest for AI supremacy, with both sides also claiming the need to keep AI under human control. Meanwhile, other commentary warned that pausing could hand China an edge in hacking, spying, and hybrid warfare, arguing that restraint without enforceable checks could be strategically dangerous. Geopolitically, the cluster points to a shift from traditional arms-control logic toward “governance-by-measurement,” where verification technologies could determine whether AI restraint is credible. The power dynamic is asymmetric: the US appears concerned about losing strategic advantage if it slows, while China is portrayed as potentially unconstrained in compute capacity, which could translate into faster model iteration and broader cyber/hybrid capabilities. A separate report described a Western push for global AI controls that echoes nuclear-era treaty thinking, suggesting a coalition-building effort that could pressure non-aligned states to accept monitoring norms. The net effect is that AI regulation is becoming entangled with national security, export controls, and technology sovereignty—meaning any “deal” will likely be judged as much by cyber deterrence and industrial leverage as by safety rhetoric. Market and economic implications are likely to concentrate in AI infrastructure, cybersecurity, and semiconductor supply chains, because verification and control efforts tend to raise compliance and instrumentation costs. If AI restraint talks gain traction, investors may price a higher probability of compute rationing, model-access limits, or cross-border monitoring requirements, which could affect demand for high-end GPUs, data-center power equipment, and cloud training capacity. Conversely, if the dominant narrative remains “no pause,” the market may continue to reward frontier-model builders and the firms supplying training and inference at scale, with spillovers into cyber-defense spending and surveillance-adjacent tooling. The Taiwan trade-policy angle also matters for supply chains: any tightening or re-routing of semiconductor-related trade could influence pricing for advanced chips and the cost of AI systems, reinforcing volatility in tech-heavy indices and related FX hedges for exporters. What to watch next is whether verification tools move from prototypes to deployable standards that both governments can trust, including auditability of training runs, provenance of model updates, and measurable compliance metrics. Executives and policymakers should monitor signals such as joint US–China technical working groups, proposals for third-party verification bodies, and any language that shifts from voluntary safety commitments to enforceable monitoring. The “global controls” push will be a second key track: watch for coalition statements, draft frameworks, and whether major powers invite Russia and Iran into governance discussions or treat them as out-of-scope. Finally, the trigger for escalation is a hardening of “supremacy” narratives—especially if cyber/hybrid-threat claims intensify—while de-escalation would look like concrete verification pilots, reciprocal transparency steps, and clearer constraints on high-risk model deployment timelines.

Geopolitical Implications

• 01

  AI arms-control analogies suggest a move toward measurable, treaty-like constraints, but verification technology could become a new domain of strategic leverage.

• 02

  Cyber and hybrid warfare concerns are likely to harden positions, making “safety” rhetoric insufficient without enforceable monitoring.

• 03

  Compute capacity narratives (e.g., China’s alleged lack of constraints) may widen the gap between model development timelines, affecting deterrence and bargaining power.

• 04

  Semiconductor and Taiwan-related trade policy may increasingly be treated as national-security infrastructure for AI, not just commerce.

Key Signals

• —Formation of US–China technical verification working groups and publication of measurable compliance metrics
• —Third-party audit proposals (or refusals) for model training and deployment logs
• —Any shift from voluntary safety commitments to enforceable monitoring language
• —Escalating rhetoric linking AI pauses to cyber/hybrid advantages
• —Trade-policy moves affecting Taiwan-linked semiconductor flows and AI hardware procurement