OpenAI has expressed concern about the rise of open-weight large language models (LLMs), particularly those developed by Chinese lab Moonshot, such as the Kimi K3 model. Dean W. Ball, OpenAI’s head of strategic futures, initially suggested that the U.S. government should create regulatory hurdles to limit the spread of these open models, arguing they could damage investment incentives for established U.S. AI companies. However, Ball later retracted his call for aggressive regulation, amid broad criticism from AI experts who argue that open-source models fuel innovation alongside proprietary ones.

Despite this retraction, reports indicate that the Trump administration considered banning these advanced Chinese LLMs from the U.S. market, mainly driven by requests from American frontier AI labs. The main concern for U.S. companies is economic: open-weight models, which can be run on independent or enterprise infrastructure, offer a cheaper alternative to proprietary models like those from OpenAI or Anthropic. This threatens the business models of those companies, which rely on returns from their expensive training investments.

Chinese open-weight models also raise security and political worries, including the potential for data sharing with the Chinese government, ideological bias, and lack of regulatory guardrails present in U.S.-approved models. Yet some U.S. companies reportedly turn to Chinese models for tasks U.S. models won’t perform, exposing a paradox in the risks versus utility debate. Experts suggest the U.S. might better safeguard its leadership in AI through targeted chip export controls rather than broadly restricting open-source AI technologies that have wide support in research and industry communities.

The broader dilemma is the economics of AI development, with neither proprietary nor open business models yet fully established. While frontier labs seek to protect their market position, advocates emphasize that open AI fosters broader innovation and shared advancement. Restrictions on open AI access could consolidate power rather than improve safety or performance, ultimately hindering global AI progress. U.S. leadership could benefit from supporting capable, less costly open models rather than attempting to suppress competitors under the guise of national security or market protectionism.