Back to today’s briefing

Story perspectives

Malicious AI Models Found, Removed from Hugging Face Hub

2/10/2025

48 8

Summary

Story summary

Two dangerous machine learning models were discovered on Hugging Face Hub, capable of creating a reverse shell on users' devices. These models cleverly bypassed detection by misusing the unsafe pickle format, remaining undetected for over eight months. Thankfully, Hugging Face has swiftly removed them and enhanced their security protocols to protect users.

1 of 1

Story summary

Two dangerous machine learning models were discovered on Hugging Face Hub, capable of creating a reverse shell on users' devices. These models cleverly bypassed detection by misusing the unsafe pickle format, remaining undetected for over eight months. Thankfully, Hugging Face has swiftly removed them and enhanced their security protocols to protect users.