CVE-2026-105755
MEDIUMDescription
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.
Is your site exposed to CVE-2026-105755?
Run a free security scan — no signup, results in seconds.
CVSS v3.1 Score
EPSS — Exploit Prediction
EPSS estimates the probability that this vulnerability will be exploited in the wild within the next 30 days. A higher score means more likely to be exploited.
Weakness Type (CWE)
Affected Products
| Vendor | Product |
|---|---|
| vllm | vllm |
References
Advisories & Patches
Other References
Frequently Asked Questions
What is CVE-2026-105755? +
How severe is CVE-2026-105755? +
What products are affected by CVE-2026-105755? +
How do I check if I'm vulnerable to CVE-2026-105755? +
Related Vulnerabilities
FreeScout is a free self-hosted help desk and shared mailbox. Prior to version 1.8.213, the `load_customer_info` action in `POST /conversation/ajax` …
A missing authentication vulnerability exists in the Altium 365 SearchService. A legacy SOAP endpoint exposes search index operations without requiring …
Horilla is a free and open source Human Resource Management System (HRMS). In 1.5.0, an insecure direct object reference in …
Horilla is a free and open source Human Resource Management System (HRMS). In 1.5.0, an insecure direct object reference in …
A vulnerability in SpiceJet’s booking API allows unauthenticated users to query passenger name records (PNRs) without any access controls. Because …
Authenticated user can bypass authorization in Ribblr - Crochet & Knitting iOS application