CVE-2025-46722

vLLM is an inference and serving engine for large language models (LLMs). In versions starting from 0.7.0 to before 0.9.0, in the file vllm/multimodal/hasher.py, the MultiModalHasher class has a security and data integrity issue in its image hashing method. Currently, it serializes PIL.Image.Image objects using only obj.tobytes(), which returns only the raw pixel data, without including metadata such as the image’s shape (width, height, mode). As a result, two images of different sizes (e.g., 30x100 and 100x30) with the same pixel byte sequence could generate the same hash value. This may lead to hash collisions, incorrect cache hits, and even data leakage or security risks. This issue has been patched in version 0.9.0.

CVSS v3 4.2 MEDIUM

4.2^/10

CVSS v3 : MEDIUM

Vector :

Exploitability : 1.6 / Impact : 2.5

Attack Vector NETWORK

Attack Complexity HIGH

Privileges Required LOW

User Interaction NONE

Confidentiality Impact LOW

Integrity Impact NONE

Availability Impact LOW

Scope UNCHANGED

References

Link	Resource
https://github.com/vllm-project/vllm/commit/99404f53c72965b41558aceb1bc2380875f5d848	Patch
https://github.com/vllm-project/vllm/pull/17378	Issue Tracking Patch
https://github.com/vllm-project/vllm/security/advisories/GHSA-c65p-x677-fgj6	Vendor Advisory

Configurations

Configuration 1 (hide)

cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*

History

24 Jun 2025, 18:12

Type	Values Removed	Values Added
New CVE

Information

Published : 2025-05-29 17:15

Updated : 2025-06-24 18:12

NVD link : CVE-2025-46722

Mitre link : CVE-2025-46722

CVE.ORG link : CVE-2025-46722

JSON object : View

Products Affected

vllm

vllm

CWE

CWE-1023

Incomplete Comparison with Missing Factors

CWE-1288

Improper Validation of Consistency within Input

{"id": "CVE-2025-46722", "cveTags": [], "metrics": {"cvssMetricV31": [{"type": "Secondary", "source": "security-advisories@github.com", "cvssData": {"scope": "UNCHANGED", "version": "3.1", "baseScore": 4.2, "attackVector": "NETWORK", "baseSeverity": "MEDIUM", "vectorString": "CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:L/I:N/A:L", "integrityImpact": "NONE", "userInteraction": "NONE", "attackComplexity": "HIGH", "availabilityImpact": "LOW", "privilegesRequired": "LOW", "confidentialityImpact": "LOW"}, "impactScore": 2.5, "exploitabilityScore": 1.6}, {"type": "Primary", "source": "nvd@nist.gov", "cvssData": {"scope": "UNCHANGED", "version": "3.1", "baseScore": 7.3, "attackVector": "NETWORK", "baseSeverity": "HIGH", "vectorString": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L", "integrityImpact": "LOW", "userInteraction": "NONE", "attackComplexity": "LOW", "availabilityImpact": "LOW", "privilegesRequired": "NONE", "confidentialityImpact": "LOW"}, "impactScore": 3.4, "exploitabilityScore": 3.9}]}, "published": "2025-05-29T17:15:21.523", "references": [{"url": "https://github.com/vllm-project/vllm/commit/99404f53c72965b41558aceb1bc2380875f5d848", "tags": ["Patch"], "source": "security-advisories@github.com"}, {"url": "https://github.com/vllm-project/vllm/pull/17378", "tags": ["Issue Tracking", "Patch"], "source": "security-advisories@github.com"}, {"url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-c65p-x677-fgj6", "tags": ["Vendor Advisory"], "source": "security-advisories@github.com"}], "vulnStatus": "Analyzed", "weaknesses": [{"type": "Primary", "source": "security-advisories@github.com", "description": [{"lang": "en", "value": "CWE-1023"}, {"lang": "en", "value": "CWE-1288"}]}], "descriptions": [{"lang": "en", "value": "vLLM is an inference and serving engine for large language models (LLMs). In versions starting from 0.7.0 to before 0.9.0, in the file vllm/multimodal/hasher.py, the MultiModalHasher class has a security and data integrity issue in its image hashing method. Currently, it serializes PIL.Image.Image objects using only obj.tobytes(), which returns only the raw pixel data, without including metadata such as the image\u2019s shape (width, height, mode). As a result, two images of different sizes (e.g., 30x100 and 100x30) with the same pixel byte sequence could generate the same hash value. This may lead to hash collisions, incorrect cache hits, and even data leakage or security risks. This issue has been patched in version 0.9.0."}, {"lang": "es", "value": "vLLM es un motor de inferencia y servicio para modelos de lenguaje grandes (LLM). En versiones desde la 0.7.0 hasta anteriores a la 0.9.0, en el archivo vllm/multimodal/hasher.py, la clase MultiModalHasher presenta un problema de seguridad e integridad de datos en su m\u00e9todo de hash de im\u00e1genes. Actualmente, serializa los objetos PIL.Image.Image utilizando \u00fanicamente obj.tobytes(), que devuelve \u00fanicamente los datos de p\u00edxeles sin procesar, sin incluir metadatos como la forma de la imagen (ancho, alto, modo). Como resultado, dos im\u00e1genes de diferentes tama\u00f1os (p. ej., 30x100 y 100x30) con la misma secuencia de bytes de p\u00edxeles podr\u00edan generar el mismo valor hash. Esto puede provocar colisiones de hash, aciertos de cach\u00e9 incorrectos e incluso fugas de datos o riesgos de seguridad. Este problema se ha corregido en la versi\u00f3n 0.9.0."}], "lastModified": "2025-06-24T18:12:30.023", "configurations": [{"nodes": [{"negate": false, "cpeMatch": [{"criteria": "cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*", "vulnerable": true, "matchCriteriaId": "08DBEEAC-7BAC-4A44-894B-5F544B5CF9D3", "versionEndExcluding": "0.9.0", "versionStartIncluding": "0.7.0"}], "operator": "OR"}]}], "sourceIdentifier": "security-advisories@github.com"}