Allocation of Resources Without Limits or Throttling Affecting vllm package, versions [0.19.0,0.31.0)


Severity

Recommended
0.0
high
0
10

CVSS assessment by Snyk's Security Team. Learn more

Threat Intelligence

Exploit Maturity
Proof of Concept
EPSS
0.41% (36th percentile)

Do your applications use this vulnerable package?

In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.

Test your applications

Snyk Learn

Learn about Allocation of Resources Without Limits or Throttling vulnerabilities in an interactive lesson.

Start learning
  • Snyk IDSNYK-PYTHON-VLLM-22556260
  • published15 Nov 2026
  • disclosed13 Nov 2026
  • creditSeonwoong Hwang

Introduced: 13 Nov 2026

NewCVE-2026-126399  (opens in a new tab)
CWE-770  (opens in a new tab)

How to fix?

Upgrade vllm to version 0.31.0 or higher.

Overview

vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs

Affected versions of this package are vulnerable to Allocation of Resources Without Limits or Throttling via the media_io_kwargs.video.num_frames field in chat completion requests, which lacks a server-enforced upper bound. An authenticated inference client can set num_frames=-1 or an arbitrarily large positive value, causing VideoMediaIO.load_base64 to decode every comma-separated JPEG frame in the request payload and combine them with np.stack, with memory consumption scaling linearly with the number of supplied frames. This bypasses the server-side frame ceiling introduced by the parent advisory (GHSA-pq5c-rjhq-qp7p), because ChatParams.with_defaults() gives the request-supplied value precedence over the engine-level limit without clamping or rejecting out-of-range values.

Note: This is only exploitable when the deployment serves a video-capable multimodal model. If --api-key is configured, a valid inference API key is required; without a configured key, no API-key middleware is enforced on the /v1/chat/completions route.

Workaround

Ensure --trust-request-mm-kwargs is not enabled (it is disabled by default in v0.31.0+). Keeping this flag disabled blocks untrusted per-request mm_processor_kwargs and media_io_kwargs before preprocessing.

References

CVSS Base Scores

version 4.0
version 3.1