CVE-2026-5241
Huggingface Transformers 5.2.0
Raw vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:HSummary
CVE-2026-5241 is a critical-severity Inclusion of Functionality from Untrusted Control Sphere (CWE-829) vulnerability in Huggingface Transformers. Its CVSS base score is 9.6 (Critical).
Operationally, exploitation aligns with the MITRE ATT&CK technique Supply Chain Compromise (T1195); ranked at the 43th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog; a public proof-of-concept is referenced.
This vulnerability is AI-related — categorised as NLP and Transformers; in the Supply Chain and Deployment risk domain.
The strongest mitigations our analysis identified map to SR-4 (Provenance) and SR-11 (Component Authenticity) — see the control section below for these in your framework.
OWASP Top 10 for Web (2025)
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2026-34084
Vulnerability Data
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…
more
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.
- CWE(s)
AI Security AnalysisAI
- AI Category
- NLP and Transformers
- Risk Domain
- Supply Chain and Deployment
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: huggingface, transformers
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
- 6 hardening rules · 6 OS baselines
V14.2.3V3.5.6V9.1.3V15.3.2
Mitigating Controls (NIST 800-53 r5) AI
Requires documented, valid provenance for components, preventing acceptance of code from outside the trusted sphere.
Implements detection and prevention of counterfeit or inauthentic components before they are integrated.
Establishes processes to identify and address supply-chain weaknesses that would allow untrusted functionality.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
Pre-acquisition integrity/authenticity checks directly prevent inclusion of untrusted code.
Supply-chain program directly governs inclusion of third-party executable code.
Contractual requirements can mandate trusted sources and integrity checks for included functionality.
Supplier risk assessment explicitly covers risks from their products and libraries.
Critical-supplier assessment reduces risk of importing executable functionality from untrusted parties.
Secure SDLC practices include dependency vetting and trusted-source policies.
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
By requiring suppliers to propagate security requirements and to disclose component provenance, the control limits the inclusion of functionality obtained from untrusted third-party sources without oversight.
Blocking domains that serve malware or untrusted scripts prevents the browser from automatically including functionality from an attacker-controlled source.
Hardening callouts derived
Configuration rules from DISA STIG baselines that bear on weaknesses of the type cited by this CVE. Each rule is shown with the relationship its mapping actually records, against the CWE it was authored against. Derived via CVE→CWE over `controls_xwalks` (authoritative rows only; rows rated `none` are excluded).
Oracle Linux 8 (1 rule)
- V-248635 Executable search paths within the initialization files of all local interactive OL 8 users must only contain paths that resolve to the system default or the user's home directory. prevents CWE-829
Oracle Linux 9 (1 rule)
- V-271847 OL 9 must be configured so that executable search paths within the initialization files of all local interactive users must only contain paths that resolve to the system default or the users home directory. prevents CWE-829
RHEL 7 (1 rule)
- V-204477 The Red Hat Enterprise Linux operating system must be configured so that all local interactive user initialization files executable search paths contain only paths that resolve to the users home directory. prevents CWE-829
RHEL 8 (1 rule)
- V-230317 Executable search paths within the initialization files of all local interactive RHEL 8 users must only contain paths that resolve to the system default or the users home directory. prevents CWE-829
RHEL 9 (1 rule)
- V-258050 Executable search paths within the initialization files of all local interactive RHEL 9 users must only contain paths that resolve to the system default or the users home directory. prevents CWE-829
Windows 10 (1 rule)
- V-220737 Administrative accounts must not be used with applications that access the Internet, such as web browsers, or with potential Internet sources, such as email. prevents CWE-829