CVE-2026-116173
Received Received - Intake

Denial of Service in vLLM via Logprob Token Limits

Vulnerability report for CVE-2026-116173, including description, CVSS score, EPSS score, affected products, exploitability, helpful resources, and attack-flow context.

Publication date: 2026-10-25

Last updated on: 2026-10-25

Assigner: VulnCheck

Description

vLLM from 0.22.0 before 0.24.0 contains a denial of service vulnerability that allows remote attackers to terminate EngineCore by bypassing logprob_token_ids limits in the Rust gRPC frontend. Attackers can send Generate or GenerateStream requests with output_logprobs enabled and more than 128 output_candidates.token_ids, triggering a fatal sampler exception that crashes the engine.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-10-25
Last Modified
2026-10-25
Generated
2026-10-25
AI Q&A
2026-10-25
EPSS Evaluated
N/A
NVD
EUVD

Affected Vendors & Products

Showing 1 associated CPE
Vendor Product Version / Range
vllm-project vllm 0.22.0

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-20 The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

vLLM versions 0.22.0 to 0.23.x contain a denial of service vulnerability in the Rust gRPC frontend. Attackers can send specially crafted requests with output_logprobs enabled and more than 128 output_candidates.token_ids. This bypasses internal limits, triggering a fatal sampler exception that crashes the EngineCore component.

Detection Guidance

Monitor vLLM engine crashes or fatal exceptions in logs when processing gRPC Generate or GenerateStream requests with output_logprobs enabled. Check for requests exceeding 128 output_candidates.token_ids in gRPC traffic.

Impact Analysis

This vulnerability allows remote attackers to terminate the vLLM engine by sending malicious requests, causing a denial of service. This disrupts services relying on vLLM for language model inference, leading to downtime and potential loss of availability for affected systems.

Mitigation Strategies

Upgrade vLLM to version 0.24.0 or later to apply the fix. If upgrading is not immediately possible, restrict access to the gRPC interface or disable output_logprobs in requests.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-116173. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart