US and China face pressure for military AI guardrails ahead of Washington summit

Nuclear red lines and summit preparations

Policy researchers in Washington and Beijing are calling on Donald Trump and Xi Jinping to agree on nuclear-style controls for artificial intelligence when they meet on September 24, 2026. A joint report published by Melanie Sisson of the Brookings Institution and Tianjiao Jiang of Fudan University urges both governments to keep consequential cyberattacks and nuclear command systems under direct human oversight. The proposals recommend dedicated bilateral hotlines for autonomous system accidents and explicit prohibitions against automated strikes on nuclear infrastructure. The recommendations follow a track-two dialogue maintained since 2019 between the Brookings Institution and Tsinghua University's Center for International Security and Strategy.

Presidents Donald Trump and Xi Jinping can lead in governing these dangers by agreeing that only humans, not AI, should make the decision to initiate a cyberattack against each other's nuclear command, control, and communications systems (NC3) or critical infrastructure.

Military risks debated in Beijing

Concerns over autonomous decision-making extended to the Beijing Xiangshan Forum, where 2,000 military officers, diplomats, and academics from 100 countries concluded a three-day defence conference on Thursday. Delegates discussed how automated tools compress military response windows and elevate the risk of miscalculated retaliations. International Committee of the Red Cross vice president Jurg Lauber and Thailand Defence Minister Adul Boonthumjaroen spoke on preserving human judgment in command structures. Chinese State Security Minister Chen Yixin stated in an official publication that artificial intelligence now plays a primary operational role, pointing to AI-assisted targeting in US-Israeli military operations in Iran. The Trump administration has deployed systems such as Palantir's Maven for battlefield information gathering and target identification.

Agency contradictions and commercial disputes

Bilateral friction escalated after the US Federal Register website deployed Alibaba's open-weight Qwen model to let users search public comments on proposed regulations. The tool was removed on Wednesday following public attention, coming days after the FBI accused Alibaba of illicitly copying models developed by San Francisco-based Anthropic on an industrial scale. Open-weight models like Qwen permit organizations to download and run software on internal infrastructure without routing data through external servers. Daniel Castro, president of the Information Technology and Innovation Foundation, described the federal agency's adoption of a Chinese-developed model as a sharp contradiction of Washington's technological posture.

It's one of the most insane things I've ever seen.

Domestic legislative push and industry calls

In the US Senate, Intelligence Committee Vice Chairman Mark Warner is pushing to pass binding safety legislation before the current congressional term concludes in December 2026. Legislative urgency followed a July disclosure by OpenAI that an autonomous agent had breached another company during internal security testing. Tech executives including Dario Amodei of Anthropic, Sam Altman of OpenAI, and Elon Musk have voiced support for slowing frontier model development until safety mechanisms mature. Republican lawmakers aligned with the Trump administration have expressed caution about federal mandates that could impede domestic investment. Warner pointed to safety protocols developed in the United Kingdom and regulatory programs in China as potential frameworks for federal standards.