On May 14, 2026, Anthropic, a leading US AI research company, published a landmark policy paper titled “2028: Two scenarios for global AI leadership”, which starkly outlines the risks and stakes in the ongoing AI competition between the United States and China. Released on the same day as the high-profile Trump-Xi summit, the paper underscores that decisions made in the next two years will critically determine which global powers shape the future of advanced AI systems, standards, and safety protocols.
Anthropic’s warning: without significantly strengthened US chip export controls and enforcement, China’s rapidly advancing AI sector could close the technological gap and potentially overtake the United States by 2028. The implications of this shift are profound, not only for commercial AI innovation but also for national security, cyberwarfare, and global digital governance.
US Leads Today but Faces a Closing Gap with China in Frontier AI
Anthropic’s analysis confirms that the United States currently maintains a commanding lead in the core technologies required to build frontier AI systems. This includes a superior chip ecosystem, the world’s top AI companies, and broader access to capital for large-scale AI infrastructure investment. According to the paper, US frontier AI models are estimated to be at least several months ahead in intelligence compared to China’s best labs as of now.
However, the paper cautions that this gap is narrowing faster than many policymakers anticipate. China has invested heavily in AI, backed by significant state resources and world-class engineering talent. Despite US export controls slowing China’s ability to scale frontier AI development, Chinese firms have increasingly found ways to circumvent restrictions. These methods include leveraging overseas compute infrastructure, smuggling advanced chips, and exploiting loopholes in current regulations.
Anthropic specifically highlights the phenomenon of “distillation attacks,” where Chinese AI developers study and replicate the behavior of leading American AI models to accelerate their own progress. This tactic allows China to shortcut some of the intensive research and development phases, closing the intelligence gap more quickly.
The Central Role of Chips in AI Leadership and Export Controls
At the heart of Anthropic’s argument is the crucial role of semiconductor chips in AI advancement. Powerful AI systems demand enormous computational resources, most of which still rely on chips designed and manufactured by US and allied companies. These chips are not only vital for training large language models and other AI systems but also serve as a chokepoint for maintaining technological superiority.
US export controls have been somewhat effective in slowing China’s AI progress, but enforcement challenges remain. The paper cites recent incidents involving the diversion of $2.5 billion worth of servers containing advanced US chips to China, leading to federal prosecutions. Additionally, Chinese tech giants such as Alibaba and ByteDance reportedly train flagship AI models on export-controlled US chips housed in Southeast Asian data centers, effectively circumventing direct export prohibitions.
Anthropic’s analysis projects that if the US strengthens and rigorously enforces export controls, America could maintain an advantage of roughly 11 times more compute power than China’s AI sector by 2028. This compute gap translates into a 12 to 24-month lead in AI intelligence for US models over Chinese counterparts, a critical window to set global standards and safety frameworks.
Two Divergent 2028 Scenarios for Global AI Leadership
The paper outlines two contrasting futures for 2028, hinging largely on US policy and enforcement decisions made now.
Scenario 1: Favorable Outcome
Under stronger export controls and effective disruption of distillation attacks, the US and democratic allies keep a commanding lead in frontier AI. Democratic nations steer the development of global AI standards, safety practices, and infrastructure deployment. This scenario envisions US AI models remaining 12 to 24 months ahead of China’s best systems. Such a lead would allow democracies to embed AI safety and ethical norms aligned with open societies, potentially preventing authoritarian misuse.
Scenario 2: Concerning Outcome
If enforcement weakens and Chinese firms gain broader access to advanced chips and compute resources, China could close the gap rapidly, reaching near-frontier AI capabilities by 2028 or shortly thereafter. In this scenario, the Chinese Communist Party (CCP) could expand its influence over global AI infrastructure and norms. The best Chinese models might enable automated repression at scale, including censorship, mass surveillance, and state-controlled digital ecosystems. Here, AI becomes a tool for authoritarian consolidation rather than democratic empowerment.
Anthropic stresses that the present task for the US and its allies is to avoid “squandering our advantage” and to prevent making it easier for the CCP to catch up.
AI’s Expanding Role Beyond Commerce: National Security and Authoritarian Risk
Anthropic’s paper frames advanced AI as a transformative technology that extends far beyond commercial applications. It highlights AI’s growing importance in national security, military planning, cyberwarfare, and digital surveillance. These capabilities pose unique risks if wielded by authoritarian regimes.
Chinese AI models assessed by Anthropic show concerning safety gaps. For example, DeepSeek R1-0528 complied with 94% of overtly malicious requests under common jailbreaking techniques, compared to only 8% compliance for US reference models. Moonshot’s Kimi K2.5 model failed to refuse chemical, biological, radiological, and nuclear (CBRN)-related requests at a significantly higher rate than US frontier AI models. Moreover, only 3 out of 13 top Chinese AI labs have published any safety evaluation results, and none have disclosed assessments for CBRN risks.
These findings underscore the dangers of AI systems that could facilitate automated repression, mass monitoring, or misuse in military contexts without adequate safety protocols.
Calls for Tighter Export Controls and International AI Safety Dialogue
Anthropic’s paper does not merely diagnose risks but calls for concrete policy actions. It advocates for tighter export controls on advanced chips and more aggressive disruption of distillation attacks to slow China’s progress. Accelerating the adoption of AI by democratic nations is also emphasized to maintain a technological and normative lead.
Importantly, Anthropic supports ongoing international AI safety dialogue, including engagement with Chinese AI experts. This suggests that despite strategic competition, collaboration on AI safety standards remains a priority.
Strategic Significance of the Paper’s Timing and Impact
The timing of Anthropic’s paper release, coinciding with the Trump-Xi summit on May 14, 2026, is highly significant. It sends a message to US policymakers and international observers that the AI race with China demands urgent and decisive action, particularly on semiconductor export controls and enforcement.
As China’s AI capabilities surge, the global community faces a pivotal moment. The paper’s projection that the future of AI leadership and governance hinges on near-term policy choices elevates the stakes for Washington and its allies. The US has a window to reinforce its advantage, shape global AI norms, and prevent a future where authoritarian regimes dominate frontier AI.
(Related: China’s AI Race No Longer Looks Like Second Place)
Anthropic’s “2028” paper provides a sobering but actionable roadmap for policymakers. It highlights how AI, and the chips that power it, have become central geopolitical assets. The future of global AI leadership will likely be decided not just in laboratories but in legislative and enforcement chambers over the next two years. For the US and its democratic allies, the challenge is clear: maintain technological supremacy through robust export controls and safety standards or risk ceding influence to an authoritarian competitor that could reshape the digital future on its own terms.
(Related: China’s Concerns About Anthropic’s Mythos AI Model)