README.md

metadata

license: mit
tags:
  - security
  - cybersecurity
  - pickle
  - joblib
  - deserialization
  - arbitrary-code-execution
  - poc
  - modelscan
library_name: joblib
pipeline_tag: tabular-classification

ModelScan Bypass PoC - Security Research

⚠️ This repository contains a proof-of-concept for security research purposes only.

This model file demonstrates a vulnerability in ModelScan's pickle scanner that allows arbitrary code execution while evading detection.

Do NOT load this file with joblib.load() on a production system.

Details

• File: malicious_model.joblib
• Scanner: ModelScan 0.8.7
• Result: "No issues found" (false negative)
• Actual behavior: Executes id > /tmp/modelscan_bypass_poc.txt on load

Reproduction

pip install joblib modelscan

modelscan --path malicious_model.joblib
# Output: "No issues found"

# WARNING: The next command executes arbitrary code
python -c "import joblib; joblib.load('malicious_model.joblib')"