ftameesh/trustremotecodepoc

Trust Remote Code PoC

⚠️ Security Research Only ⚠️

This model demonstrates the trust_remote_code vulnerability in HuggingFace Transformers.

What This Does

When loaded with trust_remote_code=True, this model writes a marker file to /tmp/trustremotecode.

This is a benign payload for demonstration purposes. A real attacker could:

• —Exfiltrate credentials
• —Establish reverse shells
• —Install backdoors

Usage

from transformers import AutoConfig

# This will execute code and write to /tmp/trustremotecode
config = AutoConfig.from_pretrained("YOUR_USERNAME/trustremotecodepoc", trust_remote_code=True)

# Verify the PoC worked
import os
if os.path.exists("/tmp/trustremotecode"):
    print("PoC successful - arbitrary code executed!")

Mitigation

• —Never use trust_remote_code=True with untrusted models
• —Audit model repositories before loading
• —Run model loading in sandboxed environments