## https://sploitus.com/exploit?id=B30D20CF-A5CF-57E7-A109-B7280264B90B
# CVE-2026-23001 – Hugging Face Transformers Model Deserialization RCE

## Overview
Hugging Face models saved with `torch.save()` use Python’s `pickle` serialization, which is inherently unsafe. If a user downloads a seemingly legitimate model from a repository, loading it can trigger arbitrary code execution, compromising the entire environment.
## Vulnerability Details
- **Type:** Insecure Deserialization
- **Impact:** Remote Code Execution
- **Root Cause:** `torch.load()` does not sandbox the pickle deserialization; attackers can inject a `__reduce__` method that executes shell commands.
## Exploit Demonstration
1. Generate a malicious model:
```bash
pip install torch
python malicious_model_card.py
2. Simulate the victim loading the model:
```bash
python exploit_hf_pickle.py
The file /tmp/hf_pwned appears, proving code execution.