Benign proof-of-concept for a ProtectAI huntr Model-File-Vulnerability submission.

poc_model.joblib executes code at joblib.load() time via a cProfile.run gadget, while modelscan (0.8.8, latest) reports "No issues found". The payload is benign: it only writes a marker file /tmp/modelscan_bypass_poc.txt to prove code execution. Do not load untrusted model files.

Repro:

modelscan -p poc_model.joblib   # No issues found
python -c "import joblib; joblib.load('poc_model.joblib')"   # executes