CVE-2026-116175: vLLM audio decompression bomb in chat completions

CVE-2026-116175 · Severity: medium · CVSS 6.5 · Published 2026-10-25

vLLM is a large language model inference engine that supports multimodal inputs including audio. Authenticated attackers can submit chat requests containing small compressed audio files that expand to enormous decoded audio data, exhausting server memory and degrading service availability. An attacker needs only valid credentials and can exploit the vulnerability either by uploading audio directly or referencing an audio file via URL.

Technologies: vLLM Project vllm. Vendors: vLLM Project.

Quick answers

Is CVE-2026-116175 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-116175 being exploited in the wild.
How severe is CVE-2026-116175?
Medium severity, with a CVSS score of 6.5, published 25 October 2026.
Which products and versions are affected?
vLLM Project vllm 0.19.0 to 0.23.x.
Is there a fix for CVE-2026-116175?
Yes: fixed in 0.24.0. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-116175 attacked?
It is attacked over the network.
Am I affected by CVE-2026-116175?
Check whether you run vLLM Project vllm at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • Fixed in: 0.24.0

Technical details

A decompression bomb vulnerability in the AudioMediaIO multimodal audio processing path fails to apply the max_duration_s decoded-duration limit when loading audio from input_audio or audio_url parameters in /v1/chat/completions requests. The load_audio() function supports duration limiting, but the chat path invokes it without passing the limit, allowing attackers to provide small compressed audio that expands to exhaustive decoded PCM samples. This is a sibling vulnerability to a prior audio decompression-bomb fix that addressed the speech-to-text path but left the chat multimodal path unpatched.

Affected products

  • vLLM Project vLLM 0.19.0 to 0.23.x

Timeline

  • 2026-10-09: disclosed
  • 2026-10-25: advisory

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.