CVE-2026-12491: Misinterpretation of Input in vllm-project vLLM
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.
AI Analysis
Technical Summary
The vulnerability in vLLM 0.11.0 arises from incorrect processing of image metadata related to EXIF orientation and PNG transparency (tRNS) during image conversion to RGB. Transparency information may be discarded or remapped improperly, causing transparent pixels to render incorrectly and input images to be distorted. This misprocessing can lead to the model misinterpreting the image content, potentially impacting the integrity of the data processed by the model. The CVSS 3.1 base score is 4.8 (medium severity), reflecting network attack vector with high attack complexity, no privileges required, no user interaction, no confidentiality impact, limited integrity impact, and low availability impact. The vendor advisory from Red Hat does not specify any patch or mitigation, and no exploits are known in the wild.
Potential Impact
The vulnerability can cause the vLLM model to misinterpret image content due to improper handling of transparency and orientation metadata. This may affect the integrity of the processed data, potentially leading to incorrect model outputs or degraded inference quality. There is no direct confidentiality impact reported. The availability impact is low, and no known exploitation has been observed.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. As of the latest Red Hat advisory, no official fix or workaround has been published. Users should monitor the vendor advisory for updates and consider restricting input images with transparency metadata until a fix is available.
CVE-2026-12491: Misinterpretation of Input in vllm-project vLLM
Description
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 v3.1
Score 4.8medium
Affected software
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
The vulnerability in vLLM 0.11.0 arises from incorrect processing of image metadata related to EXIF orientation and PNG transparency (tRNS) during image conversion to RGB. Transparency information may be discarded or remapped improperly, causing transparent pixels to render incorrectly and input images to be distorted. This misprocessing can lead to the model misinterpreting the image content, potentially impacting the integrity of the data processed by the model. The CVSS 3.1 base score is 4.8 (medium severity), reflecting network attack vector with high attack complexity, no privileges required, no user interaction, no confidentiality impact, limited integrity impact, and low availability impact. The vendor advisory from Red Hat does not specify any patch or mitigation, and no exploits are known in the wild.
Potential Impact
The vulnerability can cause the vLLM model to misinterpret image content due to improper handling of transparency and orientation metadata. This may affect the integrity of the processed data, potentially leading to incorrect model outputs or degraded inference quality. There is no direct confidentiality impact reported. The availability impact is low, and no known exploitation has been observed.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. As of the latest Red Hat advisory, no official fix or workaround has been published. Users should monitor the vendor advisory for updates and consider restricting input images with transparency metadata until a fix is available.
Technical Details
- Data Version
- 5.2
- Assigner Short Name
- redhat
- Date Reserved
- 2026-06-17T07:24:01.437Z
- Cvss Version
- 3.1
- State
- PUBLISHED
- Remediation Level
- null
- Vendor Advisory Urls
- [{"url":"https://access.redhat.com/security/cve/CVE-2026-12491","vendor":"Red Hat"}]
Threat ID: 6a3280380b89be68882feed5
Added to database: 06/17/2026, 11:08:40 UTC
Last enriched: 07/07/2026, 10:41:40 UTC
Last updated: 08/01/2026, 07:17:58 UTC
Views: 53
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.