"We're going to be discussing AI guardrails with the Chinese," Treasury Secretary Scott Bessent said during an interview with CNBC.
U.S. clearances for Nvidia H200 to ten Chinese firms
Reports emerging from the recent summit between President Donald Trump and Chinese President Xi Jinping indicate that the U.S. has cleared roughly 10 Chinese firms to buy Nvidia's H200 chips. The coverage makes two immediate qualifications: no deliveries have been made, and the approvals surfaced amid an active internal debate inside the U.S. government as it weighs competing national security and economic priorities tied to AI competition with Beijing. Decisions around chip export licenses typically fall under the Commerce Department; the Commerce Department did not respond to a request for comment.
AI at the summit: Trump, Xi, and Nvidia's Jensen Huang
AI was a key topic at the summit, and President Trump personally invited Nvidia CEO Jensen Huang to join the trip to Beijing. Public readouts differed: Chinese state-linked summaries appeared to emphasize technology coordination and AI governance discussions more than the White House's public statements, which featured little direct mention of cybersecurity or AI. In a reported exchange, Xi told Trump that "safeguarding peace and stability across the Taiwan Strait is the biggest common denominator between China and the U.S." A White House statement about the discussions did not mention Taiwan.
Scott Bessent on guardrails, models, and his knowledge of the chip approvals
On the sidelines of the summit, Bessent framed a bilateral dialogue on AI governance as an effort to keep "advanced models" out of the hands of non‑state actors, saying the two leading AI powers would begin talks around AI governance and "best practices" aimed at preventing advanced models from falling into the hands of non-state actors. He also said he expects a major "step-function jump" in upcoming large language model releases from Google’s Gemini and OpenAI. Separately, Bessent said he had no knowledge of the H200 approvals. The White House did not immediately respond to a request for comment.
Security alarm from experts and the talent flow to Chinese labs
Security analysts and former federal cyber officials consulting with ISMG warned that permitting advanced compute while simultaneously opening governance talks creates a central tension: can the U.S. hold restrictive export controls in place while pursuing limited risk-reduction dialogue with Beijing? Leah Siskind, an AI research fellow at the Foundation for Defense of Democracies, told ISMG: "The U.S. shouldn't trade export controls for dialogue." She added that, "Export controls are the single most effective constraint on PRC AI development - Chinese labs and officials say so themselves."
Siskind and other experts pointed to sharper structural differences in how U.S. frontier labs operate — citing more developed safety and security regimes, third-party red teaming, responsible scaling policies, and government engagement — and warned that Chinese labs "aren't replicating it." The concern is amplified by state-backed industrial policy, aggressive talent recruitment, and allegations of large-scale distillation of American AI models. The reporting names recent talent moves as examples: Wu Yonghui, a former Google vice president of research who helped develop Gemini and now leads research for ByteDance's AI arm, and Yao Shunyu, a former OpenAI researcher who was named Tencent Holdings' chief AI scientist. Analysts say those commercial gains could feed into military and intelligence priorities under China's military-civil fusion strategy.
What this means for technologists, policymakers, and Chinese AI firms
- Technologists and security teams: Will need to watch for changes in compute availability and model replication vectors as approvals — even before deliveries — change the calculus around access to high-end accelerators like the H200.
- Policymakers and regulators: Face a demonstrable tension between pursuing bilateral AI "guardrails" dialogue and maintaining export controls; Bessent's public separation of the two underscores that tension even as some approvals reportedly move forward.
- Chinese AI firms and labs: Stand to gain near-term access to advanced hardware if deliveries proceed, while also being subject to the diplomatic and regulatory attention that follows such approvals.
The picture that emerges from the summit is one of parallel tracks: public commitments to talk about AI governance and private movements on high‑end hardware approvals. The administration is publicly laying groundwork for "best practices," while reports say roughly ten Chinese firms have been cleared to buy Nvidia H200 chips — approvals that, according to published accounts, have not yet resulted in shipments. With the Commerce Department, the White House and the Treasury offering different lines of comment or none at all, the central unanswered question is procedural as much as strategic: will permitting compute access while initiating governance talks narrow or widen the frontier model gap the U.S. seeks to shape?
Source: GovInfoSecurity — Go-Ahead for AI Chip Sales to 10 Chinese Firms Raise Alarms
