US Treasury secretary Scott Bessent and China’s vice-premier He Lifeng agreed last weekend to establish a formal US-China dialogue on AI. The US has proposed a mutual notification mechanism about serious AI incidents, reminiscent of the Cold War nuclear hotline. These steps aim to apply guardrails to the AI race between Washington and Beijing.

Despite tensions—Trump dismissing AI risks as a 'hoax' and Beijing viewing US safety talks as containment—shared vulnerabilities exist. Both face risks of AI falling into non-state actor hands, enabling bioweapon development or cyberattacks on critical infrastructure. Agreeing on access controls, model testing, and containment could reduce dangers.

The speed of AI-driven incidents risks escalatory retaliation, making a US-style AI 'hotline' sensible for Xi to endorse. Clear rules are also needed for AI in weaponry, building on Xi’s 2024 agreement with Biden that humans must retain nuclear control. Experts since 2019 have recommended red lines: banning autonomous nuclear launch decisions, requiring human authorization for AI-driven cyberattacks on nuclear networks, and defining 'meaningful human control.' The goal is to establish safeguards before a near-miss or catastrophe from rapidly advancing AI.