GCD AI RADAR
incidents/AIR-2025-0057

Malicious Models with Embedded Malware Discovered on Hugging Face

Attack on AI3 Confirmed exploitationSupply chain

Researchers identified malicious machine learning models containing embedded reverse shell payloads hosted on the Hugging Face repository. The files utilized corrupted structures to evade scanner detection during deserialization.

Details

type
Attack on AI
An AI model, agent or AI-powered product was the thing exploited, abused or leaked.
categories
  • Supply chain
    Malicious or backdoored models, poisoned packages, unsafe deserialization and other compromised AI dependencies.
owasp llm top 10
mitre atlas
Both rows are derived from the type and categories (OWASP 2025 edition, ATLAS release 2026.09), not classified per record. How
severity
3 · Confirmed exploitation
Used against real targets; limited or unclear impact.
targets
Framework / tooling, Open-weight model
products
Hugging Face, Picklescan
incident date
2025 (approx.)
first reported
22 Apr 2025
origin
MITRE ATLAS
Derived from MITRE ATLAS (Apache 2.0, © MITRE).
classifier confidence
0.95 · published at or above 0.80; see the methodology
tags
hugging-face · malicious-models · deserialization · supply-chain
record
created 2026-10-18 06:55Z · updated 2026-10-18 06:55Z · json

Sources · 2

  1. 2025-04-22
    Malicious Models on Hugging Face
    MITRE ATLAS · atlas.mitre.org/studies/AML.CS0031
  2. 2025-04-22
    Malicious ML models discovered on Hugging Face platform
    reversinglabs.com · reversinglabs.com/blog/rl-identifies-malware-ml-model-hosted-on-hugging-face?web

Titles and links belong to their publishers. The summary above is written by GCD AI RADAR in its own words; data is CC BY-SA 4.0.

Related incidents