US, China experts call for nuclear-style safeguards on AI risks

A group of security experts from both nations has put forward a set of risk-management proposals aimed at preventing catastrophic outcomes from advanced AI systems, mirroring Cold War-era approaches to nuclear security. The framework seeks to establish clearer protocols for managing potential AI-related threats that could cross international borders or affect critical infrastructure.

Central to the proposals is the requirement that human decision-makers retain control over consequential cyberattacks, ensuring that autonomous systems cannot independently initiate significant military or civilian infrastructure operations. This safeguard addresses growing concerns about the speed of AI systems, which could theoretically act faster than human operators can respond.

The experts have also recommended establishing a direct communication channel between nations to report and manage incidents involving autonomous AI systems. Such a hotline would function similarly to Cold War-era mechanisms designed to reduce miscalculation and escalation during crises, allowing rapid diplomatic intervention when AI systems behave unexpectedly or produce unintended consequences.

The proposals reflect growing international recognition that advanced AI capabilities pose novel security challenges requiring coordinated governance. As AI systems become more sophisticated and integrated into defence and critical infrastructure, experts on both sides of geopolitical divides increasingly advocate for transparency measures and mutual safeguards to prevent accidental escalation or misuse.