The secret Trump administration battle to fight Chinese AI

Axios | Maria Curi

The Trump administration is weighing potential restrictions on Chinese open-source artificial intelligence models, a move that could solidify the market dominance of domestic firms like OpenAI and Anthropic. Officials are evaluating measures including adding Chinese AI labs to the Commerce Department's Entity List or issuing cybersecurity advisories to discourage U.S. corporate adoption of foreign technology. 

These deliberations follow the emergence of high-performing Chinese models such as Kimi, which currently compete with U.S. alternatives on both cost and capability. Internal debates persist between national security hawks advocating for strict containment and pro-competition advisers who warn that such regulations could stifle domestic innovation and lead to regulatory capture. While an outright ban remains under consideration, the administration is increasingly utilizing procurement rules and public pressure campaigns to limit the integration of Chinese platforms into the U.S. ecosystem, signaling a shift toward a more aggressive, durable approach to technological decoupling.

Comment

The potential inclusion of AI labs on the Commerce Department's Entity List represents a shift from traditional hardware-centric export controls to the regulation of algorithmic software. By targeting open-source architectures, the administration risks fragmenting the global developer ecosystem that currently relies on shared repositories for rapid innovation. This approach mirrors the logic applied to Huawei's access to semiconductor manufacturing equipment, yet it faces unique challenges due to the intangible nature of model weights and the difficulty of enforcing compliance across distributed cloud environments. Such a policy would effectively force a bifurcation of the AI development landscape, compelling firms to choose between interoperability with Chinese-origin models or alignment with U.S. regulatory standards.

To read in details click here