Anthropic Outlines Two Futures for Global AI Leadership by 2028

Two futures for artificial intelligence leadership are at stake before 2028, according to a policy essay published by Anthropic in May 2026. The company's argument centers on something less glamorous than model benchmarks: access to advanced semiconductors and the compute infrastructure required to train and run frontier AI systems.

The paper presents a branching path. In the first scenario, the US and allied democracies maintain a lead of roughly 12 to 24 months in frontier AI development, achieved by tightening export controls, disrupting what Anthropic calls "distillation attacks" by Chinese AI firms, and accelerating global adoption of American AI technology. In the second, policy gaps persist, Chinese labs remain at near-parity with US models, and China leverages cost advantages and rapid domestic deployment to extend geopolitical influence, particularly in developing markets.

The distillation attack mechanism is the part worth paying closest attention to. Anthropic says Chinese labs are currently harvesting outputs from US frontier models to replicate their capabilities without access to the underlying training infrastructure. Alongside illicit chip smuggling and offshore data centers, this is how China has, according to the paper, managed to stay competitive despite export restrictions on advanced semiconductors. The company holds that compute access is the primary bottleneck, and whoever controls that bottleneck shapes which norms get embedded in AI systems at a global scale.

On the downside scenario, Anthropic warns that authoritarian AI leadership could enable "automated repression at scale" globally, with AI norms and rules shaped by authoritarian regimes rather than democracies. The company has written that "there is a high likelihood that we will look back on 2026 as the breakaway opportunity for American AI" — framing this year as the decisive one.

The honest caveat is that Anthropic is not a disinterested analyst here. The paper's three policy recommendations — closing compute smuggling loopholes, preventing distillation attacks, and promoting global export of American AI systems — directly favor US frontier AI labs commercially. The paper also does not address whether technical countermeasures to distillation attacks are actually feasible at scale, or how American AI systems compete on price in developing markets where Chinese alternatives may be substantially cheaper. The question that matters most is not which scenario is more likely, but which specific policy levers actually shift the outcome — and on that, the paper is stronger on diagnosis than prescription.