CVE-2026-90554

vLLM before 0.28.0 Denial of Service via audio extraction

Published Sep 12, 2026 CVSS 4.0 · 6.9 Medium EPSS < 1% KEV · Not listed

Description

vLLM versions >=0.10.2 and <0.28.0 do not apply any audio decode-size or duration limit when extracting audio from video input for NanoNemotronVL models. In nano_nemotron_vl.py, _extract_audio_from_videos calls load_audio_pyav(BytesIO(video_bytes)) without the max_duration_s or max_decode_bytes parameters, so neither VLLM_MAX_AUDIO_DECODE_DURATION_S nor VLLM_MAX_AUDIO_DECODE_BYTES is enforced (unlike the direct audio upload path in AudioMediaIO). When a NanoNemotronVL model is served with use_audio_in_video=True, an attacker who supplies a small, highly compressed video as multimodal input can force the server to allocate gigabytes of memory during audio decoding, resulting in a denial of service. Fixed in vLLM 0.28.0.

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

unaffected

Version	Status	Constraints
`0.10.2`	affected	< 0.28.0
`0.28.0`	unaffected	—

Configuration 1 [-]

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

No data.

Vendor Product Versions

Vllm

Version	Status	Scheme	Platform
`[0,*]`	affected	generic	—

Vllm-project

Vllm

Version	Status	Scheme	Platform
`[0.10.2,0.28.0)`	affected	semver	—
`0.28.0`	unaffected	semver	—

Remediation

No solution or workaround provided in the CVE record

Tracking

Advisories

No advisories available

Metrics

Attack Vector Local

Attack Complexity Low

Privileges Required None

Attack Requirements None

User Interaction None

Vulnerable System Confidentiality Impact None

Vulnerable System Integrity Impact None

Vulnerable System Availability Impact High

Subsequent System Confidentiality Impact None

Subsequent System Integrity Impact None

Subsequent System Availability Impact None

Attack Vector Local

Attack Complexity Low

Privileges Required None

Scope Unchanged

Confidentiality Impact None

Integrity Impact None

Availability Impact High

User Interaction None

No CVSS v3.0

No CVSS v2

This CVE is not in the KEV list.

Exploit prediction score: < 1% (0.201% probability within 30 days)

Exploitation none

Automatable no

Technical Impact partial

References

Link	Providers
https://github.com/vllm-project/vllm/security/advisories/GHSA-936p-m5pv-vvjf
https://nvd.nist.gov/vuln/detail/CVE-2026-90554
https://www.cve.org/CVERecord?id=CVE-2026-90554
https://www.vulncheck.com/advisories/vllm-before-0.28.0-denial-of-service-via-audio-extraction

History

Sep 14, 2026 9:00 PM Scoring metrics updated View changes

Type	Values Removed	Values Added
Metrics		ssvc `{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}`

Sep 13, 2026 7:45 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

Sep 13, 2026 12:15 AM 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-2026-90554 https://www.cve.org/CVERecord?id=CVE-2026-90554
Metrics	threat_severity `None`	threat_severity `Moderate`

Sep 12, 2026 12:30 PM Description updated. Title updated. First appearance in the feed. Weaknesses updated. Affected versions updated. Affected vendors updated. Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
Description		vLLM versions >=0.10.2 and <0.28.0 do not apply any audio decode-size or duration limit when extracting audio from video input for NanoNemotronVL models. In nano_nemotron_vl.py, _extract_audio_from_videos calls load_audio_pyav(BytesIO(video_bytes)) without the max_duration_s or max_decode_bytes parameters, so neither VLLM_MAX_AUDIO_DECODE_DURATION_S nor VLLM_MAX_AUDIO_DECODE_BYTES is enforced (unlike the direct audio upload path in AudioMediaIO). When a NanoNemotronVL model is served with use_audio_in_video=True, an attacker who supplies a small, highly compressed video as multimodal input can force the server to allocate gigabytes of memory during audio decoding, resulting in a denial of service. Fixed in vLLM 0.28.0.
Title		vLLM before 0.28.0 Denial of Service via audio extraction
First Time appeared		Vllm Vllm vllm
Weaknesses		CWE-400
CPEs		cpe:2.3:a:vllm:vllm::::::::
Vendors & Products		Vllm Vllm vllm
References		https://github.com/vllm-project/vllm/security/advisories/GHSA-936p-m5pv-vvjf https://www.vulncheck.com/advisories/vllm-before-0.28.0-denial-of-service-via-audio-extraction
Metrics		cvssV3_1 `{'score': 6.2, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H'}` cvssV4_0 `{'score': 6.9, 'vector': 'CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N'}`