CVE-2025-46722: vLLM has a Weakness in MultiModalHasher Image Hashing Implementation
vllm-project · vllm
Affected: vllm-project · Exploitation: unlikely (under 1%)
Not in CyberSIXT coverage yet: no article names this CVE. This page shows the public record.
Higher than 27% of scored CVEs
DescriptionCVE.org
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.
Affected products1
From the CVE record. A product or version that is not listed is unknown, not unaffected.
- vllm-project · vllm>= 0.7.0, < 0.9.0
On the Atlas
Every published CVE is on the CyberSIXT Atlas, grouped by weakness, vendor and month.
- Exploitation
- Unlikely (under 1%). EPSS under 1%, or no score yet.
- CISA KEV
- Not listed
- Public check
- No public nuclei template
- Vendor
- vllm-project
- Published
- May 2025
CoverageNo articles yet
No CyberSIXT article names CVE-2025-46722 yet, so there is no coverage timeline, action brief or change log for it. Everything on this page comes from the public CVE record, CISA KEV, EPSS and the nuclei template index. Search the archive or see the CVEs in the news.