In a stark assessment of the global artificial intelligence landscape, Anthropic’s article “2028: Two scenarios for global AI leadership” outlines a critical competition between the United States and China for dominance in AI. Published on May 14, 2026, the analysis emphasizes the urgent need for democratic nations, particularly the US, to secure and expand their lead in AI, especially concerning access to advanced computing power, often referred to as ‘compute’. The stakes are high, as failure to maintain this advantage could allow authoritarian regimes, such as the Chinese Communist Party (CCP), to dictate the future trajectory of AI development and its global norms.
The Crossroads for AI Governance
Anthropic presents two divergent scenarios for the year 2028. The first scenario envisions a success story for democracies, where they manage to defend their compute advantage. This would be achieved through a combination of strengthened export controls, effective disruption of illicit compute access methods like distillation attacks, and accelerated AI adoption across democratic societies. In this future, democratic values would significantly shape the global AI landscape.
Conversely, the second scenario paints a grim picture of inaction. Should democratic nations fail to adequately address the challenges, China could catch up or even surpass the US in AI capabilities. This outcome would likely result in authoritarian regimes establishing the rules for AI, potentially leading to widespread automated repression and a global order dictated by non-democratic principles. The article points to the CCP’s current utilization of AI for censorship, repression, cyber-attacks, and military advancements as evidence of the threat posed by an authoritarian-led AI future.
Compute Power: The Decisive Front
The competition for AI leadership is characterized by four key fronts: intelligence, domestic adoption, global distribution, and resilience. However, Anthropic identifies compute power as the most crucial element. China’s ability to remain competitive is attributed, in part, to its access to illicit compute and its exploitation of distillation attacks, which allow for the transfer of capabilities from larger models to smaller ones.
To ensure democratic leadership in AI, Anthropic advocates for a proactive strategy. This includes closing existing loopholes in export controls to limit the flow of critical AI technology to adversaries, actively deterring distillation attacks, and promoting the widespread global export of American AI technologies. The article concludes with a powerful message: the decisions made by policymakers today will fundamentally determine whether the transformative power of AI will be guided by democratic principles or subjected to authoritarian control.
