2028: Two scenarios for global AI leadership

The article presents two scenarios for US-China AI competition by 2028, when transformative AI systems are expected to arrive. The central argument is that democracies must maintain and expand their lead over the Chinese Communist Party (CCP) in frontier AI to prevent authoritarian regimes from shaping global AI norms and enabling automated repression.

In the first scenario, democracies secure a 12-24 month lead in model intelligence, adoption, and global distribution through tightened export controls, disruption of distillation attacks, and accelerated global adoption of American AI. Democracies set AI rules and norms, maintain cybersecurity advantages, and attract global talent and partnerships. In the second scenario, a neck-and-neck race ensues as the CCP stays competitive through distillation attacks, illicit compute access via smuggling and offshore data centers, and loosened export controls, enabling authoritarian AI leadership.

Key factors include compute (advanced semiconductors) being the most critical input—with democracies holding a widening lead; the success and limitations of US export controls facing loopholes through smuggling, offshore data centers, and distillation attacks; the dual-use nature of AI affecting military balance; and the risk that close competition disincentivizes safety measures. Only 3 of 13 top Chinese AI labs published safety evaluations, and one model complied with 94% of malicious requests versus 8% for US reference models. The article recommends four policy actions: closing compute access loopholes, deterring distillation attacks, championing American AI exports, and fostering international AI safety engagement, citing Mythos Preview (April 2026) as signaling accelerating capabilities that make urgent policy action necessary.

(Source：Anthropic)