Governance Horizon Limits Traceability of Open-Weight AI Model Constraints

An investigation published on arXiv (2605.24383) examines 2,142,823 model repositories within the Hugging Face Hub to evaluate if voluntary disclosures of metadata can maintain the traceability of ethical guidelines throughout deep model lineages. The study finds that evidence of restrictions diminishes with a half-life of 1.31 derivation steps (R²=0.98). After seven generations of models, over 80% of descendants lack adequate public evidence for governance assessments, establishing a boundary known as the governance horizon. Efforts at the platform level to recover absent license metadata indicate that the design of policies, rather than enforcement alone, is crucial: designs based solely on inheritance necessitate near-complete enforcement to extend the horizon.