CVE-2026-5241

Policy Bypass in LightGlue Nested Config Resolution in huggingface/transformers

Published Jun 3, 2026 CVSS 3.1 · 9.6 Critical EPSS < 1% KEV · Not listed
Description
A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `LightGlueConfig` reads the `trust_remote_code` value from the untrusted `config.json` file and propagates it into nested `AutoConfig.from_pretrained()` calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, or persistence/backdoor deployment.

Remediation

No solution or workaround provided in the CVE record

Tracking

Sign in to view the affected projects.

Advisories

Source ID Title
Github GHSA Github GHSA GHSA-fgcw-684q-jj6r huggingface/transformers: Arbitrary Code Execution During Model Initialization in the LightGlue Model Loading Path
History
Type Values Removed Values Added
First Time appeared Huggingface
Huggingface transformers
CPEs cpe:2.3:a:huggingface:transformers:5.2.0:*:*:*:*:*:*:*
Vendors & Products Huggingface
Huggingface transformers
Metrics cvssV3_1

{'score': 7.7, 'vector': 'CVSS:3.1/AV:N/AC:H/PR:L/UI:R/S:C/C:H/I:H/A:N'}

cvssV3_1

{'score': 9.6, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H'}

Type Values Removed Values Added
References
Metrics threat_severity

None

cvssV3_1

{'score': 7.7, 'vector': 'CVSS:3.1/AV:N/AC:H/PR:L/UI:R/S:C/C:H/I:H/A:N'}

threat_severity

Important

Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'poc', 'Technical Impact': 'total'}, 'version': '2.0.3'}

Type Values Removed Values Added
Description A vulnerability in the LightGlue model loading path of huggingface/transformers version 5.2.0 allows an attacker-controlled model repository to execute arbitrary code during model initialization. The issue arises because the `trust_remote_code` parameter, intended to prevent remote code execution, is overridden by untrusted serialized configuration data in a nested code path. Specifically, when loading a LightGlue model using `AutoModel.from_pretrained()` with `trust_remote_code=False`, the `LightGlueConfig` reads the `trust_remote_code` value from the untrusted `config.json` file and propagates it into nested `AutoConfig.from_pretrained()` calls. This results in the execution of attacker-provided Python modules, even when the victim explicitly disables remote code execution. The vulnerability poses a high risk for environments such as API inference servers, research notebooks, CI/CD pipelines, and model evaluation workers, potentially leading to credential theft, lateral movement, or persistence/backdoor deployment.
Title Policy Bypass in LightGlue Nested Config Resolution in huggingface/transformers
Weaknesses CWE-829
References
Metrics cvssV3_0

{'score': 8, 'vector': 'CVSS:3.0/AV:N/AC:H/PR:N/UI:R/S:C/C:H/I:H/A:N'}

Weaknesses

Sources

Status
PUBLISHED
Assigner
@huntr_ai
Published
Updated
Sep 1, 2026, 12:04 PM
Reserved
Mar 31, 2026, 2:26 PM
Status
Modified
Published
Jun 3, 2026, 2:16 PM
Modified
Aug 28, 2026, 4:18 PM
Severity
Important
Public date
Jun 3, 2026, 12:33 PM