CVE-2026-22778
CriticalvLLM has RCE In Video Processing
Published Feb 2, 2026 · Updated Sep 10, 2026 · Also known as GHSA-4r2x-xpjr-7cvv, PYSEC-2026-565
Key Facts
Patch in the next maintenance window- CVE-2026-22778 is a critical-severity vulnerability (CVSS 9.8 v3.1) affecting Vllm, tritonserver-backend-vllm-cuda-12.9 (chainguard), tritonserver-backend-vllm-cuda-13.0 (chainguard) and 11 more products.
- It is not listed in any known-exploited vulnerabilities catalog. No public exploit code is known.
- EPSS estimates a 17% probability of exploitation in the next 30 days.
- Fixed versions are available, including tritonserver-backend-vllm-cuda-12.9 25.9.0_git20260318-r0, tritonserver-backend-vllm-cuda-13.0 25.11-r2, Vllm 0.14.1.
- Brandefense priority: patch in the next maintenance window, because of its high severity, although no exploitation is known.
CVSS v3.1
9.8
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
EPSS (30-day exploit likelihood)
17%
Higher than 96.9% of all scored CVEs
Exploited in the wild
No known
Not in a known-exploited catalog
Public exploit code
None known
Get the full CVE-2026-22778 brief as a PDF
Every affected product and fixed release, exploitation status and a first 24 hours response checklist, emailed within minutes.
Description
vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid image is sent to vLLM's multimodal endpoint, PIL throws an error. vLLM returns this error to the client, leaking a heap address. With this leak, we reduce ASLR from 4 billion guesses to ~8 guesses. This vulnerability can be chained a heap overflow with JPEG2000 decoder in OpenCV/FFmpeg to achieve remote code execution. This vulnerability is fixed in 0.14.1.
Affected Products
16 affected entries in total| Product | Affected versions | Fixed in |
|---|---|---|
| VllmVllm | >= 0.8.3, >= >= 0.8.3, < 0.14.1 | 0.14.1 |
| tritonserver-backend-vllm-cuda-12.9chainguard | See advisory | 25.9.0_git20260318-r0 |
| tritonserver-backend-vllm-cuda-13.0chainguard | See advisory | 25.11-r2 |
| rhaiis/vllm-cuda-rhel9:1772160593redhat | See advisory | n/a |
| rhaiis/vllm-cuda-rhel9:1782352847redhat | See advisory | n/a |
Patched Releases
tritonserver-backend-vllm-cuda-12.9
25.9.0_git20260318-r0
tritonserver-backend-vllm-cuda-13.0
25.11-r2
Vllm
0.14.1
Timeline
Publicly disclosed
Feb 2, 2026
Last updated
Sep 10, 2026
Weakness Type
- CWE-122 Heap-based Buffer Overflow
- CWE-209 Generation of Error Message Containing Sensitive Information
- CWE-532 Insertion of Sensitive Information into Log File
References
- advisoryRHSA-2026:19712
- advisoryRHSA-2026:30087
- advisoryRHSA-2026:30088
- advisoryRHSA-2026:30089
- advisoryRHSA-2026:3461
FAQ
Common Questions.
vLLM is an inference and serving engine for large language models (LLMs).
Affected: Vllm versions >= 0.8.3, >= >= 0.8.3, < 0.14.1; tritonserver-backend-vllm-cuda-12.9 (chainguard); tritonserver-backend-vllm-cuda-13.0 (chainguard), among others. The full list is in the CVE brief.
It is not listed in any known-exploited vulnerabilities catalog.
No, it is not currently listed in the CISA Known Exploited Vulnerabilities catalog.
No public exploit code is known.
Fixed versions are available, including tritonserver-backend-vllm-cuda-12.9 25.9.0_git20260318-r0, tritonserver-backend-vllm-cuda-13.0 25.11-r2, Vllm 0.14.1.
Brandefense priority: patch in the next maintenance window, because of its high severity, although no exploitation is known.
Is CVE-2026-22778 Running On Your Attack Surface?
Brandefense maps vulnerabilities like this one to your internet-facing assets and tells you exactly where to patch.
Request A Demo