Vulnerability Disclosure: Critical RCE in Joblib
Vulnerability
Arbitrary Code Execution (ACE) via Pickle Deserialization.
Target Format
.joblib
Description
The target application uses joblib.load() on untrusted files, which utilizes Python's pickle module. By crafting a malicious pickle payload, an attacker can trigger arbitrary command execution upon loading the model.
Proof of Concept
The included poc.joblib file triggers an execution of echo ACE_TRIGGERED when loaded.
Steps to Reproduce
- Load the model:
import joblib
joblib.load("poc.joblib")
- Observe the command execution.
Recommendation
Transition away from pickle-based serialization (e.g., joblib, pickle) to safer formats like safetensors or json.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support