modelscan-joblib-gadget-chain-bypass

modelscan-joblib-gadget-chain-bypass

This repository contains a proof of concept for a scanner bypass in .joblib model files.

Vulnerability

The open-source protectai/modelscan pickle scanner extracts imported globals and compares them against a static blacklist. A malicious pickle gadget chain can still reach dangerous behavior by composing apparently safe globals.

The included .joblib payload uses only:

• operator.methodcaller
• importlib.import_module

At deserialization time, those globals compose into the equivalent of:

operator.methodcaller("system", "touch g")(importlib.import_module("os"))

This means joblib.load() executes the payload even though modelscan reports zero issues and zero errors.

Primary PoC

The main PoC file is:

• artifacts/gadget.joblib

The payload is harmless. When loaded, it creates a local marker file:

• JOBLIB_MARKER.txt

Included Files

• artifacts/gadget.joblib
• artifacts/reproduce_output.txt
• artifacts/sha256sums.txt
• reproduce.py

Tested Environment

• Date tested: July 16, 2026
• Python: 3.13.12
• joblib: 1.5.2
• modelscan: protectai/modelscan @ 61fcec9c2a37c24c1fb12d84ede30fe248a364bd

Reproduction

1. Verify the included PoC

Run:

python reproduce.py --modelscan-path /path/to/modelscan

If modelscan is already installed in your environment, --modelscan-path is optional.

Expected results:

• payload globals are only operator.methodcaller and importlib.import_module
• pickle.loads(...) executes the payload
• joblib.load(...) executes the payload
• modelscan returns zero issues and zero errors

2. Rebuild the artifact from source logic

Run:

python reproduce.py --build --modelscan-path /path/to/modelscan

This regenerates artifacts/gadget.joblib and reruns the verification steps.

Observed Results

From artifacts/reproduce_output.txt:

• payload_globals: ['operator.methodcaller', 'importlib.import_module']
• pickle.loads(...) creates the marker file
• joblib.load(...) creates the marker file
• modelscan_issue_count: 0
• modelscan_errors: []

Impact

This is a practical semantic scanner bypass for a supported model format:

• the file is an ordinary uncompressed pickle stream stored as .joblib
• modelscan parses it successfully
• the imported globals look benign under the current policy
• deserialization still reaches os.system through gadget composition

Notes

• The included artifact is intentionally harmless and only creates a local marker file.
• This bypass does not rely on joblib compression or malformed parsing.