Summary
Anthropic has published what Matthew Berman describes as the most concrete geopolitical essay he has seen from a major AI lab — a policy paper titled “2028: Two Scenarios for Global AI Leadership” that explicitly frames the US-China AI competition as an existential question to be resolved within the next few years. The essay argues that American and allied leadership in AI is essential to prevent authoritarian governments, specifically naming the Chinese Communist Party, from using advanced AI systems for repression, mass surveillance, and global power projection. Berman reviews the paper in full, highlighting where he agrees with Anthropic’s risk analysis and where he diverges on the proposed solutions.
Among the essay’s key claims: access to advanced chips (not data or researchers) is the single most important ingredient for frontier AI development; US export controls on Nvidia and other American chip makers have been effective; and Chinese labs have closed the gap primarily through talent, loophole exploitation, and what Anthropic terms large-scale distillation attacks — using outputs from American models to train competing Chinese systems. The paper also references PLA deployment of DeepSeek models for military coordination, including swarm unmanned vehicle operations and cyber offense.
Berman contextualizes Anthropic’s 2028 focus as likely tied to anticipated breakthroughs in self-improving AI and compressed R&D cycles in semiconductors and biotech, areas the essay hints at without fully elaborating. He notes a sharp contrast between Anthropic’s pro-export-control stance and Nvidia CEO Jensen Huang’s more skeptical position, and acknowledges his own perspective as an American observer while engaging seriously with the essay’s arguments.
📺 Source: Matthew Berman · Published May 15, 2026
🏷️ Format: News Analysis
