CVE-2026-12491: medium-severity vulnerability in vllm-project vLLM
Vllm: vllm: image exif rotation & png trns transparency not normalized, causing mismatch between model input and expectations
Published · Updated
E-mail alert when it lands in KEV, gets a public PoC or its score goes up.
13Vexday Risk Score
No sign of exploitation. No public exploitation artifact known so far.
ssvc Trackcvss 4.8epss 0.2%
A flaw was found in vLLM, an open-source library for large language model inference. This vulnerability arises from improper handling of image metadata, specifically EXIF orientation and PNG transparency (tRNS) data, during image processing. When images are converted to RGB, transparency information may be implicitly discarded or remapped, leading to unexpected rendering of transparent pixels and distortion of input content. This can result in the model misinterpreting image content, potentially affecting the integrity of processed data.
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:L
Affected products
Red Hat · Red Hat AI Inference ServerRed Hat · Red Hat Enterprise Linux AI (RHEL AI) 3Red Hat · Red Hat OpenShift AI (RHOAI)vllm-project · vLLMRelated CVEs — vllm-project vLLM
In the same product, most dangerous first.
CVE-2026-23536HIGHFeast: unauthenticated arbitrary file readEPSS 2.5%CVE-2024-11831MEDIUMNpm-serialize-javascript: cross-site scripting (xss) in serialize-javascriptEPSS 1.1%CVE-2026-18948CRITICALFeast: feast: unsafe dill deserialization of registry-stored udfs — rce on feature server and registry serverEPSS 1.1%CVE-2026-23537CRITICALFeast: unauthenticated arbitrary file writeEPSS 1.0%CVE-2026-0603HIGHOrg.hibernate/hibernate-core: hibernate: information disclosure and data deletion via second-order sql injectionEPSS 1.0%CVE-2026-0990MEDIUMLibxml2: libxml2: denial of service via uncontrolled recursion in xml catalog processingEPSS 1.0%