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.
References
| Link | Resource |
|---|---|
| https://github.com/vllm-project/vllm/security/advisories/GHSA-936p-m5pv-vvjf | Vendor Advisory |
| https://www.vulncheck.com/advisories/vllm-before-0.28.0-denial-of-service-via-audio-extraction | Third Party Advisory |
Configurations
History
24 Sep 2026, 20:28
| Type | Values Removed | Values Added |
|---|---|---|
| First Time |
Vllm vllm
Vllm |
|
| CPE | cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:* | |
| References | () https://github.com/vllm-project/vllm/security/advisories/GHSA-936p-m5pv-vvjf - Vendor Advisory | |
| References | () https://www.vulncheck.com/advisories/vllm-before-0.28.0-denial-of-service-via-audio-extraction - Third Party Advisory |
Information
Published : 2026-09-12 13:16
Updated : 2026-09-24 20:28
NVD link : CVE-2026-90554
Mitre link : CVE-2026-90554
CVE.ORG link : CVE-2026-90554
JSON object : View
Products Affected
vllm
- vllm
CWE
CWE-400
Uncontrolled Resource Consumption