Skip to main content

CVE-2025-46722 - VLLM Image Hash Collision Vulnerability

CVE ID : CVE-2025-46722
Published : May 29, 2025, 5:15 p.m. | 3 hours, 19 minutes ago
Description : 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.
Severity: 4.2 | MEDIUM
Visit the link for more details, such as CVSS details, affected products, timeline, and more...

About

Kenya Education Network CERT(KENET-CERT) is a Cybersecurity Emergency Response Team and Co-ordination Center operated by the National Research and Education Network of Kenya. KENET-CERT coordination center promotes awareness on cybersecurity incidences as well as coordinates and assists member institutions in responding effectively to cyber security threats and incidences. KENET-CERT works closely with Kenya's National CIRT coordination center (CIRT/CC) as a sector CIRT for the academic institutions. KENET promotes use of ICT in Teaching, Learning and Research in Higher Education Institutions in Kenya. KENET aims to interconnect all the Universities, Tertiary and Research Institutions in Kenya by setting up a cost effective and sustainable private network with high speed access to the global Internet. KENET also facilitates electronic communication among students and faculties in member institutions, share learning and teaching resources by collaboration in Research and Development of Educational content.