US, China experts outline military AI safeguards

Security experts from the United States and China have proposed a set of military artificial intelligence safeguards, likening them to nuclear-level precautions. These recommendations, published on September 9 by Melanie Sisson and Tianjiao Jiang, aim to prevent potential escalation from autonomous systems causing error, malfunction, unauthorized actions, or misinterpretations.

The proposal emphasizes retaining human authority for initiating AI-enabled cyber operations against nuclear command, control, and communications systems, as well as other critical infrastructure. Sisson and Jiang advocate for explicit red lines prohibiting AI from independently deciding to use nuclear weapons or attacking nuclear command systems.

They also suggest developing a shared definition of "meaningful human control" to avoid potential misunderstandings during crises. The recommendations stem from a US-China dialogue on AI and national security, with both nations endorsing a previous commitment in November 2024 to maintain human control over nuclear weapons decisions.

However, the effectiveness of these safeguards may hinge on whether officials actively utilize the proposed dedicated US-China military hotline for incidents involving autonomous systems. Despite not publicly adopting the recommendations, China has integrated AI into its Foreign Ministry's Department of Arms Control, acknowledging its potential impact on strategic stability and the risk of miscalculation.

The United States, meanwhile, lacks a dedicated arms-control body for AI, with policy spread across various agencies. As the two nations remain competitive in AI and advanced computing, they exhibit caution in accepting restrictions that could hinder technological advancement or diminish strategic advantages.