---
title: "Don't Trust the Label: License Laundering in AI Supply Chains"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.20300v1"
permalink: "https://feed.astralor.com/items/itm_2bbeae63656fc17eb921bc28"
published: "2026-07-22T15:48:04.000Z"
score: 78
channel: "tech"
tags: ["AI Supply Chain","GitHub","Hugging Face","Open Source Compliance","License Laundering"]
---

# Don't Trust the Label: License Laundering in AI Supply Chains

## AI commentary

> This paper gave me chills—think slapping an MIT license on your model means you're safe? Actually, tons of unlabeled artifacts in the chain let obligations slip through like sand. A must-read for anyone running model aggregators or training products on open datasets—don't wait for a copyright notice to hit your inbox.

## At a glance

This study is the first to quantify license laundering in AI supply chains, tracing 232,270 dataset→model→application chains. It finds that 62.3% of chains pass through at least one artifact with no declared license, and every obligation-bearing license category falls below 7% end-to-end survival, while the permissive category reaches 95.1%. The research reveals widespread stripping or replacement of license obligations across platforms, offering actionable recommendations for practitioners, rights holders, and platform owners.

## Key points

- Traced 232,270 complete dataset→model→application supply chain chains
- Identified two forms of license laundering: unlabeled artifacts acquire downstream labels, and license categories are swapped during redistribution
- 62.3% of chains pass through at least one artifact with no declared license, and obligation-bearing licenses have <7% end-to-end survival
- Permissive licenses like MIT achieve 95.1% survival, indicating obligations are often circumvented

## Full text

AI artifacts move through a multi-platform supply chain, spanning datasets and models on Hugging Face and applications on GitHub. While each artifact carries a license whose obligations should propagate through redistribution, no study has yet measured whether those obligations survive the chain or are stripped and replaced as artifacts move downstream. We trace 232,270 dataset$\rightarrow$model$\rightarrow$application chains and quantify two forms of license laundering: when artifacts with no declared license acquire definitive labels downstream, and when one declared license category replaces another during redistribution. We find that 62.3% of chains pass through at least one artifact with no declared license (concentrated in a small set of foundational datasets), and that every obligation-bearing license category falls below 7% end-to-end survival while the Permissive category reaches 95.1%. Based on these findings, we provide actionable recommendations for practitioners, model publishers, rights holders, and platform owners.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_2bbeae63656fc17eb921bc28