Browse / Model Evasion / CVE-2026-12491
vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations
- Class
- Model Evasion (filed by a model that read the advisory)
- Technique
- Other weaknesses
- Affected
- vllm (PyPI)
- CVSS
- CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:L
- EPSS
- 0.2% chance of exploitation in 30 days
How it works, in short
Images with EXIF rotation or PNG transparency tricks are preprocessed differently than humans see them, so the model receives hidden or altered content. (Our summary, written by a model. The source text is below.)
Details
Summary
Issue 1: EXIF orientation not normalized → The image orientation processed by the model differs from how humans view it, introducing interpretation bias.
Issue 2: PNG tRNS not explicitly flattened before converting to RGB → After conversion, transparent/semi-transparent pixels are rendered unexpectedly, making otherwise subtle overlay elements visible and distorting the input content. (This attack is similar to AlphaDog: RGBA handling is already correct in vLLM, but since tRNS permits RGB images, the correct processing path isn’t taken.)
Issue 3 : Pillow only loads the first frame when loading APNG or GIF files.
---
Root Cause
- Rotation: After opening an image,
ImageOps.exif_transposeis not called to normalize EXIF orientation. - Transparency: Only RGBA→RGB is flattened with a background; PNGs carrying
tRNSinP/L/RGB + tRNSand other non-RGBA modes take theimage.convert("RGB")path, which implicitly discards/remaps transparency semantics.
---
Affected Code
https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L77-L84
https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L37-L43
https://github.com/vllm-project/vllm/blob/16b37f3119918c1e5a39f303e0d0892c65c07a90/vllm/multimodal/image.py#L26-L34
> Current state: ImageOps.exif_transpose is not used. (Although the rescale_image_size function (https://github.com/vllm-project/vllm/blob/main/vllm/multimodal/image.py#L14) exists and includes a transpose parameter, I’ve found that it doesn’t seem to be called anywhere outside the test directory.)
> Call order: _convert_image_mode runs first; if the conditions are met, convert_image_mode is called.
>
> Issue: Only the “RGBA → RGB” path is explicitly flattened. P, L, or RGB with tRNS all fall back to image.convert("RGB"). For PNGs that include tRNS, convert("RGB") directly produces 24-bit RGB, leading to:
>
> * P mode: The transparent index becomes an actual RGB color (often black, white, or an undefined background), so transparency is lost.
> * L/LA and RGB + tRNS: convert("RGB") doesn’t composite against a chosen background first, so elements that relied on transparency to be hidden or softened become solid.
Impact & Scope
- Impact: Pixels the model sees can diverge from operator expectations (due to orientation or transparency handling), potentially altering downstream reasoning.
- Scope: The image I/O and mode-conversion paths in
vllm/multimodal/image.py. The existing RGBA→RGB flattening is correct; the issues center on missing EXIF normalization and non-RGBAtRNSnot being explicitly composited.
Case
EXIF: http://qiniu.funxingzuo.top/exif_orient_180.jpg tRNS: http://qiniu.funxingzuo.top/hello.png
Fix
A fix for this vulnerability was merged here: https://github.com/vllm-project/vllm/pull/44974
References
- https://github.com/vllm-project/vllm/security/advisories/GHSA-8jr5-v98p-w75m
- https://nvd.nist.gov/vuln/detail/CVE-2026-12491
- https://github.com/vllm-project/vllm/pull/44974
- https://github.com/vllm-project/vllm/commit/cf1c90672404548aa3bc51f92c4745576a65ee26
- https://access.redhat.com/security/cve/CVE-2026-12491
- https://bugzilla.redhat.com/show_bug.cgi?id=2489786
- https://github.com/advisories/GHSA-8jr5-v98p-w75m
- https://github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-3406.yaml
- https://github.com/vllm-project/vllm
- https://pypi.org/project/vllm
Source: GitHub Advisory Database, used under CC BY 4.0.