Offensive AI Security› Model Supply-Chain Attacks›Weight and Adapter Tampering: Malicious Merges and Detecting Behavioral Drift

In this series

Model Supply-Chain Attacks

1The Artifact Threat Model: What a Downloaded Checkpoint Can Do at Load TimeResearch
2Deserialization and Loader Attacks: Pickle Code Execution and Why safetensors HelpsResearch
3Weight and Adapter Tampering: Malicious Merges and Detecting Behavioral DriftResearch
4Distribution and Typosquatting: Model-Hub Trust and Dependency ConfusionResearch
5Signing and Provenance: Model Signatures, SLSA Attestations, and Verification at LoadResearch

← All research

HostExpert

Murali Chillakuru

Press play to begin the walkthrough.

0:00

15:51

Model Supply-Chain Attacks · 3 / 5

Weight and Adapter Tampering: Malicious Merges and Detecting Behavioral Drift

Back