CVE-2026-71281

Hugging Face peft's LoRA-GA and CorDA initialization modules (src/peft/tuners/lora/corda.py lines ~102 and ~163, and src/peft/tuners/lora/loraga.py line ~101) call torch.load on config-specified cache/covariance files without weights_only=True, bypassing peft's own safe-loading wrapper used elsewhere in the codebase.

CVSS v3 8.8 HIGH

8.8^/10

CVSS v3 : HIGH

Vector :

Exploitability : 2.8 / Impact : 5.9

Attack Vector NETWORK

Attack Complexity LOW

Privileges Required NONE

User Interaction REQUIRED

Confidentiality Impact HIGH

Integrity Impact HIGH

Availability Impact HIGH

Scope UNCHANGED

References

Link	Resource
https://github.com/huggingface/peft
https://github.com/huggingface/peft/blob/main/src/peft/tuners/lora/corda.py

Configurations

No configuration.

History

23 Oct 2026, 21:10

Type	Values Removed	Values Added
Summary		(es) Los módulos de inicialización de LoRA-GA y CorDA de peft de Hugging Face (src/peft/tuners/lora/corda.py líneas ~102 y ~163, y src/peft/tuners/lora/loraga.py línea ~101) llaman a torch.load() en archivos de caché/covarianza especificados en la configuración sin weights_only=True, eludiendo el propio envoltorio de carga segura de peft utilizado en otras partes de la base de código. Debido a que torch.load() sin weights_only=True realiza una deserialización completa de pickle, la carga de un archivo de caché o covarianza malicioso (por ejemplo, una caché de LoRA-GA o CorDA compartida/descargada) resulta en ejecución de código arbitrario.

Information

Published : 2026-08-05 13:24

Updated : 2026-10-23 21:10

NVD link : CVE-2026-71281

Mitre link : CVE-2026-71281

CVE.ORG link : CVE-2026-71281

JSON object : View

Products Affected

No product.

CWE

CWE-502

Deserialization of Untrusted Data

{"id": "CVE-2026-71281", "cveTags": [], "metrics": {"ssvcV203": [{"source": "134c704f-9b21-4f2e-91b3-4a467353bcc0", "ssvcData": {"id": "CVE-2026-71281", "role": "CISA Coordinator", "options": [{"exploitation": "none"}, {"automatable": "no"}, {"technicalImpact": "total"}], "version": "2.0.3", "timestamp": "2026-08-05T15:51:18.837851Z"}}], "cvssMetricV31": [{"type": "Secondary", "source": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c", "cvssData": {"scope": "UNCHANGED", "version": "3.1", "baseScore": 8.8, "attackVector": "NETWORK", "baseSeverity": "HIGH", "vectorString": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H", "integrityImpact": "HIGH", "userInteraction": "REQUIRED", "attackComplexity": "LOW", "availabilityImpact": "HIGH", "privilegesRequired": "NONE", "confidentialityImpact": "HIGH"}, "impactScore": 5.9, "exploitabilityScore": 2.8}]}, "affected": [{"source": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c", "affectedData": [{"vendor": "huggingface", "product": "peft", "versions": [{"status": "affected", "version": "0", "versionType": "semver", "lessThanOrEqual": "0.19.1"}], "programFiles": ["src/peft/tuners/lora/corda.py", "src/peft/tuners/lora/loraga.py"], "defaultStatus": "unaffected"}]}], "published": "2026-08-05T13:24:52.723", "references": [{"url": "https://github.com/huggingface/peft", "source": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c"}, {"url": "https://github.com/huggingface/peft/blob/main/src/peft/tuners/lora/corda.py", "source": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c"}], "vulnStatus": "Deferred", "weaknesses": [{"type": "Secondary", "source": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c", "description": [{"lang": "en", "value": "CWE-502"}]}], "descriptions": [{"lang": "en", "value": "Hugging Face peft's LoRA-GA and CorDA initialization modules (src/peft/tuners/lora/corda.py lines ~102 and ~163, and src/peft/tuners/lora/loraga.py line ~101) call torch.load on config-specified cache/covariance files without weights_only=True, bypassing peft's own safe-loading wrapper used elsewhere in the codebase."}, {"lang": "es", "value": "Los m\u00f3dulos de inicializaci\u00f3n de LoRA-GA y CorDA de peft de Hugging Face (src/peft/tuners/lora/corda.py l\u00edneas ~102 y ~163, y src/peft/tuners/lora/loraga.py l\u00ednea ~101) llaman a torch.load() en archivos de cach\u00e9/covarianza especificados en la configuraci\u00f3n sin weights_only=True, eludiendo el propio envoltorio de carga segura de peft utilizado en otras partes de la base de c\u00f3digo. Debido a que torch.load() sin weights_only=True realiza una deserializaci\u00f3n completa de pickle, la carga de un archivo de cach\u00e9 o covarianza malicioso (por ejemplo, una cach\u00e9 de LoRA-GA o CorDA compartida/descargada) resulta en ejecuci\u00f3n de c\u00f3digo arbitrario."}], "lastModified": "2026-10-23T21:10:00.157", "sourceIdentifier": "309f9ea4-e3e9-4c6c-b79d-e8eb01244f2c"}