By weakness (CWE)

CWE-115: related vulnerabilities

CVEs classified under CWE-115. Understanding the weakness class helps prioritize systemic fixes over one-off patches.

1 published vulnerability

• CVE-2026-12491MEDIUM 4.8

  vLLM, a widely-used open-source library for running large language models, has a flaw in how it processes images. When the library converts images to a standard format (RGB), it does not properly preserve image metadata like EXIF orientation information or transparency data (PNG tRNS chunks). This mishandling can cause transparent pixels to be rendered incorrectly or remapped unexpectedly, distorting the original image content that the model sees. The result is that the language model may misinterpret what's in the image, compromising data integrity.