Security research PoC for responsible disclosure through Huntr's Model File Vulnerability program.
The payload is benign. Loading the model writes a marker file at /tmp/joblib_mfv_marker.txt.
1python -m venv .venv
2. .venv/bin/activate
3pip install -r requirements.txt
4
5rm -f /tmp/joblib_mfv_marker.txt
6python reproduce.py
7test -f /tmp/joblib_mfv_marker.txt && echo "marker created"
1modelscan -p . -r json --show-skipped
2picklescan -p . -g
This is distinct from generic pickle/joblib deserialization and from already-public compressed-joblib or inline NumPy object-array bypasses.