CVE-2026-12491 vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations
Improper image metadata handling in vLLM causes model input distortion, potentially leading to incorrect inference results.
- Exploited in the wild
- Not known
- Login needed
- Not needed
- Attackable
- Over the internet
- Victim action needed
- Not needed
- Fix
- Available
What is CVE-2026-12491?
CVE-2026-12491 is a medium-severity vulnerability in Red Hat AI Inference Server, published on 17 Jun 2026. An attacker can exploit it over the internet without logging in, with no action from the victim.
Official 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.
Weakness: CWE-115
Which versions are affected by CVE-2026-12491?
3 open source packages ship an affected version, including vllm.
| Package | Affected | Fixed in |
|---|---|---|
vllm PyPI | From 0.11.0 to 0.23.0 | None |
vllm PyPI | From 0.11.0 | 0.24.0 |
vllm PyPI | From 0.11.0 | 0.24.0 |
How do I fix CVE-2026-12491?
Upgrade vllm to 0.24.0 or later.
Is CVE-2026-12491 being exploited?
There is no known exploitation of CVE-2026-12491 in the wild and no public exploit code.
Sources
Also known as GHSA-8jr5-v98p-w75m, MINI-33jp-mfhh-vjp9, MINI-364f-6mjv-mhrh, MINI-g835-hm8q-2fpg, PYSEC-2026-3406.