Go to homepageCyberStrike

Menu

Star on GitHub · 3.4k
Sign inDownload

CVE-2026-100654

6.5 · MEDIUM

vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not…

Published Sep 26, 2026vllmCWE-129EPSS 0.31%

Overview

CVE-2026-100654 is a medium-severity vulnerability affecting vllm vllm. It was published on September 26, 2026 and has a CVSS 3.1 base score of 6.5 (MEDIUM).

This vulnerability has a CVSS 3.1 base score of 6.5, rated MEDIUM. It can be exploited remotely over the network. Some level of privileges is required for exploitation.

Technical description

vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).

Remediation

Check the references section for vendor advisories and patches from vllm. Update vllm to the latest patched version. If immediate patching is not possible, review the CVSS vector to understand the attack surface and apply compensating controls such as network segmentation or access restrictions.

Affected products

VendorProductVersionsStatus
vllmvllm>= 0, < 0.29.0Affected

References

GitHubgithub.com/vllm-project/vllm/security/advisories/GHSA-v5gm-qgmv-gc6cRefwww.vulncheck.com/advisories/vllm-before-0.29.0-denial-of-service-via-out-of-range-stop-token-ids

How CyberStrike helps

Frequently asked questions

What is CVE-2026-100654?

CVE-2026-100654 is a medium-severity vulnerability affecting vllm vllm. It was published on September 26, 2026 and has a CVSS 3.1 base score of 6.5 (MEDIUM).

How severe is CVE-2026-100654?

This vulnerability has a CVSS 3.1 base score of 6.5, rated MEDIUM. It can be exploited remotely over the network. Some level of privileges is required for exploitation.

How do I fix or remediate CVE-2026-100654?

Check the references section for vendor advisories and patches from vllm. Update vllm to the latest patched version. If immediate patching is not possible, review the CVSS vector to understand the attack surface and apply compensating controls such as network segmentation or access restrictions.

How can CyberStrike help with CVE-2026-100654?

CyberStrike is an open-source, AI-augmented pentesting engine. Rather than flag a version number, it tests whether an issue like CVE-2026-100654 is actually reachable in your own applications and APIs, then verifies each finding before reporting it. That tells you if the weakness is a real, exploitable risk in your environment or a false positive you can safely deprioritize.

Related MEDIUM CVEs from 2026

View all
Browse by year2026202520242023202220212020201920182017201620152014201320122011201020092008200720062005200420032002200120001999
Watch the demo →