A First Look at Model Supply Chain: From the Risk Perspective

The rapid proliferation of publicly available artificial intelligence models on platforms such Hugging Face has formed a complex model supply chain, where base models are continually transformed, e.g., by fine-tuning, quantization, and merging, into new derived models. Despite its critical importance for reuse, provenance, and risk management, this supply chain remains poorly characterized at scale.

We construct the first comprehensive model supply chain based on models on Hugging Face as of June 25, 2025, which consists of 1.82 million models as well as 0.54 million dependency relations. Then, we conduct the first systematic study to characterize the usage, evolution, quality, and risk of this model supply chain. Finally, we provide actionable implications for model maintainers, consumers, platform designers, and researchers to foster this new direction in the era of AI.