Full Breakdown
U.S.–China Experts Push Nuclear-Style Safeguards for AI Amid Rising Tensions
By Drooid · · How we work
Core Proposal: AI Safeguards Modeled on Nuclear Arms Control
Security experts in the U.S.–China AI dialogue released recommendations treating advanced AI systems like nuclear weapons. The proposals call for red lines barring AI from autonomously accessing nuclear command, control and communications (NC3) networks, require “meaningful human control” over AI-enabled cyberattacks on critical infrastructure, and establish a dedicated hotline for autonomous-AI incidents. Published the week before the Reuters report dated September 17, 2026, they aim to inform the high-level AI talks slated for September 24, 2026, in Washington.
Background & Context
The dialogue, jointly convened by the Brookings Institution and Tsinghua University’s Center for International Security and Strategy, has met regularly since 2019. The latest proposals build on a principle endorsed by former President Joe Biden and President Xi Jinping in November 2024 that humans must retain ultimate authority over nuclear weapon use. Brookings senior fellow Melanie Sisson warns that an autonomous system could malfunction or be compromised, leaving Washington or Beijing with only minutes to determine intent.
Official Statements & Responses
- China’s disarmament ambassador Shen Jian told the U.N. in Geneva that military AI must remain under human control to preserve strategic stability.
- U.S. official Bessent is scheduled to meet Chinese Vice Premier He Lifeng in New York ahead of the September 24 summit.
- Chinese officials have placed AI within the Foreign Ministry’s Department of Arms Control, treating it alongside nuclear weapons and missiles.
Criticism & Opposition
Carla Freeman of Johns Hopkins SAIS questions the practicality of the existing U.S.–China crisis-communication hotline, noting Chinese officials declined to answer U.S. calls during the “spy balloon” incident in February 2023. She also points to China’s top-down decision-making hierarchy, which can delay low-ranking officers from communicating during fast-moving AI incidents.
Conflicting Reports & Gaps
Sources differ on the effectiveness of a dedicated AI hotline. Proponents argue a direct line could allow rapid clarification of accidental AI behavior, while Freeman’s analysis of past hotline failures suggests bureaucratic delays and mistrust may limit its utility. The United States lacks a single, centralized AI arms-control office, with policy dispersed across the National Security Council, State Department and Pentagon, whereas China has formalized AI oversight within its arms-control bureaucracy.
What’s Next
The scheduled September 24 meeting between President Donald Trump and President Xi Jinping will be the first government-level forum to address the AI safeguards outlined by the expert panel. In the days leading up to that summit, Treasury Secretary Bessent’s talks with Vice Premier He Lifeng are expected to cover both “open-weight” and “closed-weight” AI models, as well as related trade and rare-earth issues. The outcome will shape whether the proposed nuclear-style safeguards become a formal component of U.S.–China AI governance.