The
rapid expansion of open-source artificial intelligence (AI), particularly large
language models (LLMs), is reshaping the global governance of AI. Open-source
approaches promise broader access and faster innovation. However, they also
raise significant governance challenges, particularly regarding transparency
and accountability in model development. Software licensing has traditionally
served as an important governance mechanism in open-source ecosystems. It
structures disclosure requirements and enables community oversight. In the
emerging open-source AI landscape, however, the relationship between licensing
and transparency remains uncertain. This paper examines how licensing
strategies shape the diffusion and governance of open-source AI. It focuses on
China’s expanding open-source AI ecosystem. Chinese AI firms have increasingly
released open-weight models under permissive licenses. This approach diverges
from the closed-source strategies adopted by leading U.S. firms. The analysis
draws on a dataset of approximately 15,000 fine-tuned models derived from major
base models on Hugging Face between 2023 and 2025. Using this dataset, the
paper examines how licensing choices influence global adoption and transparency
practices. The results show that permissive licensing significantly accelerates
the global diffusion of AI models. This effect is particularly strong for
Chinese model families. However, permissive licenses are not associated with
greater transparency. Most models disclose little information about training
data, development pipelines, or evaluation processes. These findings suggest
that open-source licensing facilitates rapid ecosystem expansion. However, it
cannot ensure transparency or accountability on its own. Effective AI
governance will therefore require regulatory frameworks that directly address
disclosure and oversight alongside open-source distribution.