CVE-2026-105760

vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion

Published Oct 5, 2026 CVSS 3.1 · 5.3 Medium EPSS < 1% KEV · Not listed

Description

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0.

Analysis

Technical impact

vLLM version prior to 0.30.0 allows callers to set large fps and max_frames values in media_io_kwargs for the GLMGA video backend, creating a huge frame-index list that consumes disproportionate CPU time and memory, reflecting vulnerabilities identified as CWE-400 and CWE-770. The resulting resource exhaustion can cause the inference service to become unresponsive, effectively denying service to legitimate users.

Affected Systems

vLLM is an inference engine for large language models. The vulnerability applies to all releases of the vllm-project's vllm package before the 0.30.0 release. Any deployment using earlier versions and referencing the GLMGA video backend is at risk.

Risk and Exploitability

The CVSS score of 5.3 indicates moderate severity. The EPSS score of <1% indicates a very low likelihood of exploitation, and the vulnerability is not listed in CISA KEV. The likely attack vector is remote, via an API call that supplies a crafted media_io_kwargs payload. An attacker can trigger the vulnerability by submitting a request with oversized fps or max_frames values to the GLMGA backend, causing resource exhaustion in the shared media-loading executor.

Affected Products

Default status is the baseline for the product, each version can override it (e.g. patched versions marked unaffected).

Vendor Product Default status Versions

vllm-project

vllm

affected

Version	Status	Constraints
`>= 0.23.0rc2, < 0.30.0`	affected	—

Configuration 1 [-]

cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*

No data.

Vendor Product Versions

Vllm-project

Vllm

Version	Status	Scheme	Platform
`[0.23.0rc2,0.30.0)`	affected	semver	—

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

Upgrade vllm to version 0.30.0 or later.
If upgrading is not immediately possible, limit or validate fps and max_frames values for the GLMGA backend to prevent excessive resource use.
Continuously monitor CPU and memory usage of the inference service for anomalous spikes that may indicate exploitation.

Generated by OpenCVE AI on October 7, 2026 at 08:09 UTC.

Tracking

Advisories

Source	ID	Title
Github GHSA	GHSA-58v5-2m8f-94pr	vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion

Metrics

No CVSS v4.0

Attack Vector Network

Attack Complexity Low

Privileges Required None

Scope Unchanged

Confidentiality Impact None

Integrity Impact None

Availability Impact Low

User Interaction None

No CVSS v3.0

No CVSS v2

This CVE is not in the KEV list.

Exploit prediction score: < 1% (0.297% probability within 30 days)

Exploitation none

Automatable yes

Technical Impact partial

References

Link	Providers
https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7
https://github.com/vllm-project/vllm/pull/54935
https://github.com/vllm-project/vllm/releases/tag/v0.30.0
https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr
https://nvd.nist.gov/vuln/detail/CVE-2026-105760
https://www.cve.org/CVERecord?id=CVE-2026-105760

History

Oct 16, 2026 8:45 PM First appearance in the feed. Affected versions updated. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm Vllm vllm
CPEs		cpe:2.3:a:vllm:vllm::::::::
Vendors & Products		Vllm Vllm vllm

Oct 6, 2026 3:30 PM Scoring metrics updated View changes

Type	Values Removed	Values Added
Metrics		ssvc `{'options': {'Automatable': 'yes', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}`

Oct 6, 2026 12:15 PM Weaknesses updated. Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
Weaknesses		CWE-770
References		https://nvd.nist.gov/vuln/detail/CVE-2026-105760 https://www.cve.org/CVERecord?id=CVE-2026-105760
Metrics	threat_severity `None`	threat_severity `Moderate`

Oct 6, 2026 1:15 AM First appearance in the feed. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm-project Vllm-project vllm
Vendors & Products		Vllm-project Vllm-project vllm

Oct 5, 2026 11:15 PM Description updated. Title updated. Weaknesses updated. Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
Description		vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0.
Title		vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion
Weaknesses		CWE-400
References		https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7 https://github.com/vllm-project/vllm/pull/54935 https://github.com/vllm-project/vllm/releases/tag/v0.30.0 https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr
Metrics		cvssV3_1 `{'score': 5.3, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L'}`