type: incident · created: 2026-08-03 · updated: 2026-08-18 · tags: [] · confidence: high · affected_sectors: [] · au_impact: true
Hugging Face Diffusers Flaws (FaceHugger) Could Let Model Repositories Execute Arbitrary Code
Summary
Three high-severity security flaws disclosed in Hugging Face's Diffusers library could allow crafted model repositories to stealthily execute arbitrary code on machines that load it. Named FaceHugger, the vulnerabilities bypass trust_remote_code, the safeguard designed to stop unreviewed code from running.
Key Details
- Date: 2026-08-03
- Product: Hugging Face Diffusers library
- Vulnerability name: FaceHugger
- Severity: High (3 flaws)
- Impact: Arbitrary code execution via crafted model repositories
- Safeguard bypassed:
trust_remote_code— the mechanism designed to prevent unreviewed code from running - Source: The Hacker News
- Reliability: Tier 2/4 — Established cyber journalism
Significance
With Hugging Face becoming the "GitHub of the AI era", these flaws pose significant supply chain risk to AI/ML pipelines. Any organisation loading models from Hugging Face using the Diffusers library could be vulnerable to silent code execution, undermining the trust model that underpins the open-source AI ecosystem.
Related
- Cyber Digest 2026 08 04