The U.S. wants to contain China’s AI. Silicon Valley keeps using it (restofworld.org)

Recent discussions among AI leaders highlight the intricate dynamics of AI development between the U.S. and China. Tarun Chhabra, Anthropic's chief national security officer, raised concerns about Chinese firms like Zhipu distilling capabilities from advanced American models, framing it as a national security issue. This narrative, however, was complicated by announcements from Mira Murati’s Thinking Machines, which openly acknowledged utilizing Chinese models like DeepSeek-V3 and Moonshot AI for its own foundation models. This contradiction illustrates the increasingly intertwined nature of AI advancements, where knowledge exchange occurs across borders regardless of political tensions. The implications of this trend are significant for the AI/ML community and global governance. As frontier models become more capable, the competition now extends to who can produce the most effective "teacher" models that inform others. The practice of model distillation, once perceived primarily as a tool for efficiency, now intersects with geopolitical concerns over intellectual property and competitive advantage. Notably, Apple's integration of Chinese models into its AI offerings signals that these models have achieved a level of sophistication that legitimizes their use in global AI infrastructure. This evolving landscape poses a challenge for policymakers, as concerns over technological boundaries blur amid the reciprocal flow of innovation between the U.S. and China, necessitating a re-evaluation of regulatory approaches in this domain.