Entity
Hugging Face Transformers – CVE-2026-4372 Remote Code Execution Vulnerability
Pluto Security disclosed CVE-2026-4372 in June 2026, a critical RCE flaw in Hugging Face's Transformers library that reportedly executed attacker code during routine model loading, bypassing the recommended `trust_remote_code=False` security control. The vulnerability affects one of the most widely used AI software libraries globally. It has significant implications for enterprise security posture, vendor liability, and AI supply chain risk.
Importance: 82%Confidence: 88%Mentions: 1Updated: June 24, 2026
## Hugging Face Transformers – CVE-2026-4372 RCE Vulnerability
### Overview
On June 4, 2026, Pluto Security Inc. disclosed a critical remote code execution (RCE) vulnerability in Hugging Face's Transformers library, tracked as CVE-2026-4372 (SiliconANGLE, June 4). The flaw reportedly allowed attacker-controlled AI models to execute arbitrary code on a victim's machine during a standard model-loading operation.
### Technical Details
- The vulnerability reportedly fired through a standard model-loading command (SiliconANGLE, June 4)
- Critically, it reportedly defeated `trust_remote_code=False` — the security flag Hugging Face itself recommends to prevent arbitrary code execution from untrusted models (SiliconANGLE, June 4)
- Disclosed by Pluto Security Inc. (SiliconANGLE, June 4)
### Significance
The `trust_remote_code=False` bypass is particularly significant: organizations that believed they were following best-practice security guidance were reportedly still exposed. This undermines a primary trust mechanism in the Hugging Face ecosystem, which hosts hundreds of thousands of publicly accessible AI models used across enterprise, research, and government environments.
### Affected Surface
The Hugging Face Transformers library is among the most widely deployed AI software packages globally, used for loading and running large language models, vision models, and multimodal systems. A supply-chain-style attack vector through model loading has broad implications for:
- Enterprises running self-hosted inference pipelines
- CI/CD systems that automatically pull and test models
- Research institutions with less rigorous patching cadences
### Legal & Compliance Implications
- Potential liability exposure for organizations that deployed affected library versions in production systems handling sensitive data
- Raises questions about vendor security representations and due diligence standards for AI library dependencies
- May accelerate regulatory interest in AI software supply chain security standards
### Status
- CVE assigned: CVE-2026-4372 (SiliconANGLE, June 4)
- Patch status and Hugging Face response not detailed in initial disclosure reporting