vllm-project vLLM up to 0.27.x Audio Extraction nano_nemotron_vl.py _extract_audio_from_videos allocation of resources

Summaryinfo

A vulnerability classified as problematic was found in vllm-project vLLM up to 0.27.x. Impacted is the function _extract_audio_from_videos of the file nano_nemotron_vl.py of the component Audio Extraction. Executing a manipulation can lead to allocation of resources. The identification of this vulnerability is CVE-2026-90554. The attack may be launched remotely. There is no exploit available. Upgrading the affected component is advised.

Detailsinfo

A vulnerability was found in vllm-project vLLM up to 0.27.x. It has been declared as problematic. This vulnerability affects the function _extract_audio_from_videos of the file nano_nemotron_vl.py of the component Audio Extraction. The manipulation with an unknown input leads to a allocation of resources vulnerability. The CWE definition for the vulnerability is CWE-770. The product allocates a reusable resource or group of resources on behalf of an actor without imposing any restrictions on the size or number of resources that can be allocated, in violation of the intended security policy for that actor. As an impact it is known to affect availability. CVE summarizes:

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.

The advisory is shared for download at github.com. This vulnerability was named CVE-2026-90554 since 09/12/2026. The exploitation appears to be easy. The attack can be initiated remotely. No form of authentication is required for a successful exploitation. There are known technical details, but no exploit is available. The MITRE ATT&CK project declares the attack technique as T1499.

Upgrading to version 0.28.0 eliminates this vulnerability.

VulDB is the best source for vulnerability data and more expert information about this specific topic.

Productinfo

Type

• Project Management Software

Vendor

• vllm-project

Name

• vLLM

Version

• 0.1
• 0.2
• 0.3
• 0.4
• 0.5
• 0.6
• 0.7
• 0.8
• 0.9
• 0.10
• 0.11
• 0.12
• 0.13
• 0.14
• 0.15
• 0.16
• 0.17
• 0.18
• 0.19
• 0.20
• 0.21
• 0.22
• 0.23
• 0.24
• 0.25
• 0.26
• 0.27

Website

• Product: https://github.com/vllm-project/vllm/

CPE 2.3info

• 🔒
• 🔒
• 🔒

CPE 2.2info

• 🔒
• 🔒
• 🔒

CVSSv4info

CVSSv3info

CVSSv2info

Exploitinginfo

Threat Intelligenceinfo

Countermeasuresinfo

Timelineinfo

Sourcesinfo

Entryinfo

VulDB is the best source for vulnerability data and more expert information about this specific topic.

Discussion