US Advisor Proposes China AI Sanctions, Shifting Tech Battle to Software

The proposition by Scott Bessent, a key economic advisor to Donald Trump, to levy sanctions on China for alleged AI model "theft" marks a significant escalation in the US-China tech rivalry. This strategic pivot extends the conflict from the hardware domain of semiconductors to the core intellectual property of AI software itself. It directly responds to the rapid emergence of high-performance Chinese open-weight models from firms like 01.AI and Zhipu AI, which are increasingly challenging the performance benchmarks set by US leaders like OpenAI and Anthropic, thereby threatening America’s perceived leadership in the generative AI space. The mechanism of "theft" primarily involves training Chinese models on the outputs of premier US APIs, a process that allows developers to bypass billions in R&D costs and proprietary data collection, fundamentally altering the unit economics of model development. This creates clear winners—Chinese enterprises and developers gaining access to near-state-of-the-art models at a fraction of the cost—and losers, namely the US AI giants whose API-driven business models are directly threatened. This forces a strategic recalculation for firms like Google and OpenAI, whose competitive moat is now proven to be more permeable than previously assumed. The forward-looking implications suggest a potential splintering of the global AI ecosystem. In the next 6-12 months, expect US firms to aggressively roll out new watermarking and synthetic data detection techniques while lobbying for policy protection. If tariffs or other sanctions are implemented, they could precipitate a long-term bifurcation, forcing nations to align with either a US or Chinese AI stack. The critical variable is enforcement: proving a model was trained on another’s output is technically challenging, making any sanctions regime difficult to police effectively. This trajectory suggests the era of a single, open global AI research community is ending.