AI Supply Chain Galaxy: 3D Visual Analytics for License Compliance

An analysis of 908,449 machine-learning models hosted on Hugging Face found that 55.46% carry compliance risks or metadata conflicts, according to a paper introducing a new 3D auditing tool called AI Supply Chain Galaxy [1][2]. The paper, posted to arXiv on 15 June 2026, describes the AI ecosystem as a “highly interconnected supply chain” where traditional compliance tools cannot keep pace with multi-hop dependency networks [1][2]. The authors present AI Supply Chain Galaxy (AISCG), an interactive visual analytics system that maps models into a three-dimensional spatial layout and integrates a rule-based compliance engine [2]. It supports multi-scale exploration, from global community detection to path-aware lineage tracing [2]. The researchers used AISCG to audit the Hugging Face Hub, a platform that hosts over 12,000 open-source machine-learning demos and has collaborated with arXiv since 2022 to embed interactive demos directly alongside papers [6][7]. The audit uncovered distinct risk patterns. Adapter derivations lacked licenses 56.67% of the time, and 8.05% of fine-tuned models exhibited “license drift,” where downstream terms diverged from the original license [1][2]. A case study on the Llama model family demonstrated how AISCG lets analysts trace inherited restrictive terms across deep topological networks [2]. The Llama architecture has been widely reused; Meta’s comparable Llama 3.1 model consumed roughly ten times the computing power of DeepSeek’s V3 model, according to company claims [9]. The paper argues that visual provenance tools reduce the cognitive load of compliance auditing as model supply chains grow more complex [2]. Hugging Face’s existing infrastructure already links models, demos, and papers. A Space can be associated with an arXiv paper by including a paper link in its README file or by linking an intermediate model on the Hub [8]. The AISCG system builds on this ecosystem by adding spatial analytics and automated rule checks [2]. The authors did not provide a release timeline for the tool.