Policy paper outlines two scenarios for US-China AI competition by 2028, emphasizing compute export controls and distillation attack prevention

A new paper presents two 2028 scenarios for US-China AI competition: one in which democracies establish a commanding lead in model intelligence and adoption through tightened export controls and disruption of distillation attacks, and another in which the CCP becomes competitive at the near-frontier because policymakers do not act.

AI labs in China have remained close to the frontier by exploiting loopholes in US export control policies and carrying out large-scale distillation attacks (efforts that harvest innovations from US models to mimic their capabilities), according to the paper.

The paper argues that if the US and its allies act now to tighten compute controls and address distillation, it may be possible to lock in a 12–24 month lead in frontier capabilities by 2028, which would be enormously advantageous and support engagement with AI experts in China on safety.