CVE-2026-105755

Received Received - Intake

Flash Late-Interaction Scoring in vLLM

Publication date: 2026-10-05

Last updated on: 2026-10-05

Assigner: GitHub, Inc.

Description

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at the /score and /rerank endpoints derives each worker's query_key value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim's identifier can overwrite the cached query embedding so the victim's documents are scored against the attacker's query, and shared use counters can also cause a late-interaction cache-miss error. This issue is fixed in version 0.30.0.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published

2026-10-05

Last Modified

2026-10-05

Generated

2026-10-06

AI Q&A

2026-10-06

EPSS Evaluated

N/A

NVD

CVE-2026-105755

Affected Vendors & Products

Vendor	Product	Version / Range
vllm-project	vllm	< 030.0

Helpful Resources

Exploitability

CWE

KEV

CWE ID	Description
CWE-639	The system's authorization functionality does not prevent one user from gaining access to another user's data or record by modifying the key value identifying the data.

Attack-Flow Graph

Executive Summary

The vulnerability in vLLM before version 0.30.0 allows an attacker to manipulate the flash late-interaction scoring at the /score and /rerank endpoints by reusing a victim's X-Request-Id header value. This can overwrite cached query embeddings, causing the victim's documents to be scored against the attacker's query.

Detection Guidance

Detection involves checking if your vLLM instance is running a version prior to 0.30.0. Use commands like 'pip show vllm' or 'vllm --version' to verify the installed version. Monitor logs for cache-miss errors or unexpected query scoring behavior.

Impact Analysis

This vulnerability could lead to incorrect scoring of documents, potentially exposing sensitive information or causing denial of service due to cache-miss errors. It may also allow attackers to infer private data based on scoring results.

Compliance Impact

This vulnerability could potentially impact compliance with GDPR and HIPAA by enabling unauthorized access to or manipulation of user data through cache poisoning or query embedding overwrites. If exploited, it may lead to improper handling of sensitive information, violating data protection requirements under these regulations.

Mitigation Strategies

Upgrade vLLM to version 0.30.0 or later immediately. Restart services to apply changes. Review and restrict access to /score and /rerank endpoints if exposed to untrusted users.

Hi! I’m here to help you understand CVE-2026-105755. Ask me anything about the vulnerability, its impact, or mitigation strategies.

0/70