CVE-2026-69147
Description
A flaw was found in vLLM, an inference and serving engine for large language models. An attacker can exploit this by submitting specially crafted video requests that force the use of the PyNvVideoCodec GPU decoder. This bypasses the engine's static GPU memory reservation, allowing the attacker to exhaust shared GPU memory. The consequence is a denial of service (DoS), leading to request failures or worker crashes.
Statement
A MODERATE severity denial-of-service flaw was found in vLLM, the inference and serving engine shipped in Red Hat AI Inference Server and in Red Hat OpenShift AI (RHOAI) vLLM-based components. A remote attacker with privileges to submit requests to a video-capable GPU deployment where the PyNvVideoCodec GPU decoder is installed can send specially crafted video requests that set media_io_kwargs.video.video_backend to pynvvideocodec, forcing use of the PyNvVideoCodec decoder even when the startup configuration selected a software decoder. Because the engine reserves decoder memory only according to its static configuration, the request-selected decoder's CUDA context, decoder surfaces, and decoded-frame allocations are not removed from the KV-cache budget, allowing the attacker to exhaust shared GPU memory and cause request failures, worker crashes, or denial of service. Red Hat rated this flaw MODERATE with CVSS 3.1 score 6.5 (AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H), consistent with the upstream rating; the impact is limited to availability, with no confidentiality or integrity impact. Affected Red Hat AI Inference Server and OpenShift AI components are marked DEFER because Red Hat does not file trackers for MODERATE severity flaws with a CVSS score below 7.0; the fix is deferred to a future product update, and the residual risk is denial of service on deployments that accept video requests on GPU hardware with PyNvVideoCodec installed. Deployments that do not process video requests or do not have the PyNvVideoCodec decoder installed are not exposed to this attack path.
Mitigation
Mitigation for this issue is either not available or the currently available options do not meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to widespread installation base, or stability. The upstream vLLM project addressed this flaw in version 0.28.0, while fixes for Red Hat AI Inference Server and Red Hat OpenShift AI are currently deferred. Exposure can be reduced by not installing the PyNvVideoCodec decoder in vLLM serving images, since exploitation requires that backend to be present.
Common Vulnerability Scoring System (CVSS) Score Details
Info alert:Important note
CVSS scores for open source components depend on vendor-specific factors (e.g. version or build chain). Therefore, Red Hat's score and impact rating can be different from NVD and other vendors. Red Hat remains the authoritative CVE Naming Authority (CNA) source for its products and services (see Red Hat classifications).
CVSS v3 Score Breakdown
| Red Hat | NVD | cve.org | |
|---|---|---|---|
| Base Score | 6.5 | N/A | 6.5 |
| Attack Vector | Network | N/A | Network |
| Attack Complexity | Low | N/A | Low |
| Privileges Required | Low | N/A | Low |
| User Interaction | None | N/A | None |
| Scope | Unchanged | N/A | Unchanged |
| Confidentiality | None | N/A | None |
| Integrity Impact | None | N/A | None |
| Availability Impact | High | N/A | High |
Vector
Red Hat: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
cve.org: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
Understanding the Weakness (CWE)
Availability
Technical Impact: DoS: Resource Consumption (CPU); DoS: Resource Consumption (Memory); DoS: Resource Consumption (Other)
When allocating resources without limits, an attacker could prevent other systems, applications, or processes from accessing the same type of resource. It can be easy for an attacker to consume many resources by rapidly making many requests or causing larger resources to be used than is needed.
Frequently Asked Questions
Not sure what something means? Check out our Security Glossary.
Want to get errata notifications? Sign up here.