AI model supply chain
Also called: ML supply chain · AI supply chain
The AI model supply chain is everything a deployed model depends on: training data, base models, fine-tunes, weight files, model hubs, conversion tools and the libraries that load them. A compromise anywhere in it can put backdoored weights or malicious code into production.
How it works
Risks include model files in formats that execute code when loaded, impersonated or compromised hub accounts, poisoned datasets, tampered conversions and quantisations, and vulnerable serving frameworks. Provenance is often thin: few models ship signed weights or full data lineage.
How to defend
Prefer safe weight formats such as safetensors, pull models from verified publishers, pin and hash what you deploy, scan model files, and record provenance.
Real cases 1 story
INFO
Research · UIUC Kang Lab
PersistBD: backdoors survive the fine-tuning meant to remove themPersistBD is a method showing that a backdoor planted in a third-party model can survive a developer's benign fine-tuning, with the reinforcement learning stage afterwards often preserving or strengthening the hidden behavior.