Per CNBC citing Reuters sources, the two governments are preparing a scheduled, frontier-model-specific negotiation — a structural shift from the export-ban era, with limits that matter as much as the meeting itself.
The Diplomatic Arc — Key Dated Events
May 2026 — Trump-Xi Summit
US and China unveil a joint AI “best practices” protocol. Treasury Secretary Scott Bessent tells CNBC the US can engage on AI because “we are in the lead.”
Through 2025–2026 — Unilateral Controls Era
US restricts advanced Nvidia chip exports to China; gates access to leading frontier models; floats restrictions on Chinese open-weight models including Kimi K3. Z.AI stands up a gigawatt data center on a domestic chip stack.
Unnamed sources tell Reuters that the US and China plan bilateral AI talks for September 2026, focused on regulating risks from rival frontier models. Timing and agenda remain unconfirmed.
~September 24, 2026 — Xi’s Planned US Visit
AI talks expected to precede Xi Jinping’s anticipated visit to the United States. Neither the visit nor the talks are formally confirmed as of the CNBC report date.
According to CNBC, citing Reuters sources, the United States and China are planning bilateral AI-specific talks for September 2026 — a direct outcome of the May 2026 Trump-Xi summit. The discussions are expected to take place before Chinese President Xi Jinping’s anticipated visit to the US around September 24. On the American side, Treasury Secretary Scott Bessent is set to lead. The stated purpose: confronting how the two governments regulate the risks posed by increasingly capable frontier AI models on both sides. Three important hedges belong in the same sentence as the news: the reporting is based on unnamed sources; the talks are planned, not held; and the agenda, scope, and any binding outcomes are entirely unknown.
The September meeting, if it occurs, builds on a protocol on AI “best practices” the two countries unveiled at the May summit — itself a thin but real first step toward shared norms. Bessent’s public framing at the time was explicit: the US can afford to negotiate on AI because, in his words, “we are in the lead.” That self-assessment deserves a bracket. Recent developments — Kimi K3, a near-frontier Chinese open-weight model that has drawn enough US concern to prompt active discussion of restrictions, and Z.AI’s gigawatt data center running on an entirely domestic chip stack — complicate any simple claim to a durable, widening advantage.
The institutional framing is also worth noting. Bessent leads from Treasury, not the Pentagon or the NSC. That is the same official behind a proposed FINRA-style AI safety watchdog for frontier models — a regulatory architecture that frames AI risk as economic and strategic statecraft as much as a defense question. The choice of who sits across the table, and from which department, signals how the US intends to frame the negotiation: less a military standoff, more a structured economic competition with jointly managed risk at the frontier.
The key insight: A scheduled, AI-specific bilateral negotiation between the US and China is a structural shift in the relationship — from one side acting unilaterally through export bans and access gates, to two governments attempting to co-manage a risk they both now recognize as shared. That shift in mode matters, regardless of what the talks produce.
The Structural Read
The dominant frame for US-China AI policy over the past two years has been unilateral control: the US moves, China responds, and the moves compound. Export restrictions on advanced Nvidia chips. Access gates on leading US frontier models. Proposed restrictions on Chinese open-weight models inside US infrastructure. Each of these is a move one side makes alone, designed to widen or preserve an asymmetric gap. A scheduled bilateral negotiation is a different instrument entirely — it acknowledges that both sides have frontier-model risk they cannot fully manage in isolation, and that a shared protocol, however weak, has value.
The Business Engineer lens here is the Permission Layer: the structural idea that governments — not developers, not deployers — ultimately set the terms on which the most capable AI systems are allowed to operate, scale, and cross borders. What is changing is that the Permission Layer is beginning to be negotiated between governments, not just asserted by one. That is a maturation of the dynamic, even between rivals — and it tracks directly with the analysis in the AI’s Geopolitical Chokepoint framework, which mapped how control of the frontier infrastructure layer — chips, models, data — becomes a geopolitical instrument as capabilities scale.
The Geopolitical Fencing of the Frontier
Treaties govern the closed frontier — not the open one
This is the structural paradox at the center of any US-China AI protocol. Two governments can align on safety norms for the large, controllable, API-gated frontier labs — GPT-class, Gemini-class systems that require massive centralized infrastructure and are visible to regulators. What they cannot govern is the open-weight edge: models like Kimi K3 that anyone can download, fine-tune, and run without asking permission from either government. A bilateral deal most naturally covers the closed, leading tier. The open tier keeps diffusing regardless. The Geopolitical Fencing of the Frontier framework traces exactly this split — and why the fence, however carefully built, has a structural gap at the open-weight seam.
A second structural observation concerns the posture Bessent’s leadership signals. Treasury-led AI diplomacy, framed as negotiating from strength, positions frontier AI as strategic economics — closer to trade and financial statecraft than to arms control. That framing has consequences for what gets on the table and what stays off. Defense-adjacent capabilities, open-weight proliferation, and compute self-sufficiency — the areas where China’s domestic progress is most visible — are the hardest to govern through an economic statecraft lens. Meanwhile, the White House’s gating of US frontier model access and the developer-level token-share shift toward non-US models on platforms like OpenRouter are both signals that the competitive landscape is already more fluid than a single “we are in the lead” framing captures.
Three Implications
IMPLICATION 1 — FROM UNILATERAL CONTROL TO CO-GOVERNANCE
Export bans and model-access restrictions are tools one side deploys without the other’s consent. A bilateral safety protocol is a different instrument: it requires both parties to acknowledge a shared risk they cannot fully contain alone. Even a weak protocol signals that the dominant unilateral-control posture of 2024–2025 has reached a ceiling — and that both governments now see value in at least the appearance of coordinated frontier-risk management. That is a real shift in the relationship’s operating mode, independent of what any agreement actually produces.
IMPLICATION 2 — THE FRAME IS SET BY WHO LEADS, AND FROM WHAT POSTURE
Treasury leadership frames this as economic statecraft from perceived strength — not arms control from parity. That framing determines which risks get on the agenda (commercial deployment risk, catastrophic-misuse norms) and which stay off (open-weight proliferation, domestic compute self-sufficiency, military applications). It also means the US is implicitly staking the negotiation on the durability of its lead — a lead that Kimi K3’s developer traction and Z.AI’s infrastructure self-sufficiency are actively contesting. The posture is coherent; the premise deserves scrutiny.
IMPLICATION 3 — PROTOCOLS GOVERN WHAT GOVERNMENTS CAN SEE AND REACH
Any bilateral AI safety agreement will most naturally bind the parties’ leading, centralized, API-gated frontier labs — the systems that are visible, attributable, and infrastructure-dependent. Open-weight models diffuse outside that governance perimeter by design. The practical effect: a US-China protocol could meaningfully shape norms for the closed, highest-capability tier while the open tier continues to propagate through developer ecosystems without asking either government’s permission. That structural gap is not a failure of the talks; it is a feature of the technology. Policymakers and enterprises building AI strategy around regulatory clarity should account for it.
Business Engineer Framework
The Permission Layer — How Governments Set the Terms of the Frontier
The Permission Layer framework maps how regulatory and geopolitical actors determine which AI systems are allowed to operate, scale, and cross borders — and why control of that layer is becoming as strategically significant as control of the compute or model layer itself. The US-China bilateral talks are the Permission Layer entering a negotiation phase: the question is no longer only what each government permits unilaterally, but what two rival governments are willing to co-permit at the frontier. The Map of AI situates the Permission Layer within the full nine-layer stack — from silicon to sovereign governance — and shows where the leverage points actually sit.
Per CNBC citing Reuters sources — with all the uncertainty that sourcing carries — the US and China are moving from the unilateral-control era of AI competition into something structurally different: a scheduled, frontier-specific negotiation in which both governments acknowledge shared risk at the top of the capability stack. Treasury leadership, “best practices” framing, and the “we are in the lead” posture all signal how the US intends to approach the table; the open-weight surge from Chinese labs signals how durable that lead actually is. A planned talk is not an agreement, a protocol is not enforcement, and the open frontier does not wait for either. But two rival superpowers agreeing to sit down specifically on frontier-model risk — ahead of a state visit, led by the official designing a domestic AI watchdog — is a real marker of how central AI governance has become to the relationship itself.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
