gcdRadar
archive/Patch now/AIR-2026-0808
closedclosed 6 Jul 2026Patch now2 Disclosed vulnerabilityAI

CVE-2026-54234: vLLM remote denial of service via invalid recovered token reinjection

vLLM

A vulnerability in vLLM allows a crafted multi-request speculative workload or gRPC sequence to trigger an out-of-vocabulary token error. This crashes the inference engine with a GPU device-side assertion failure.

disclosed CVENo confirmed exploitation, so this record is tracked as a disclosed vulnerability and not counted as an incident. It moves into the counts if CISA or the vendor confirms exploitation. Why

CVSS
7.5
EPSS
0.01 · top 50% · 2026-11-13
fix status
Fix status unknown
first report
6 Jul 2026
last update
17 Jul 2026

Timeline · 2

  1. 6 Jul 2026CVE-2026-54234: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
  2. 17 Jul 2026CVE-2026-54234: vLLM has Remote DoS via Invalid Recovered Token Reinjection

Sources · 2

Advisories and databases

  1. 2026-07-06
  2. 2026-07-17
    CVE-2026-54234: vLLM has Remote DoS via Invalid Recovered Token Reinjection
    GitHub Advisory Database · github.com/advisories/GHSA-8wr5-jm2h-8r4f

Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.

Classification

AI role
Attack on AI
categories
Infrastructure vulnerability
severity
2 · Disclosed vulnerability
fix status
Fix status unknown
targets
Framework / tooling
owasp llm top 10
LLM03:2025 Supply Chain
mitre atlas
no technique for these categories
dates
happened 17 Jul 2026 · first reported 6 Jul 2026
origin
GitHub Advisory Database
Vulnerability record from NVD / the GitHub Advisory Database. This product uses the NVD API but is not endorsed or certified by the NVD.
record
confidence 0.95 · created 2026-10-18 07:00Z · updated 2026-10-18 10:04Z

Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned

Copy a brief

This record as a short Markdown brief, for a ticket, a chat or a coding agent.

the brief as text
# CVE-2026-54234: vLLM remote denial of service via invalid recovered token reinjection

https://gcdradar.vercel.app/incidents/cve-2026-54234-vllm-remote-denial-of-service-via-invalid-rec · AIR-2026-0808 · closed 2026-07-06

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-07-06 (happened 2026-07-17)
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-54234
- CVSS: 7.5

## Summary

A vulnerability in vLLM allows a crafted multi-request speculative workload or gRPC sequence to trigger an out-of-vocabulary token error. This crashes the inference engine with a GPU device-side assertion failure.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-07-17: CVE-2026-54234: vLLM has Remote DoS via Invalid Recovered Token Reinjection (https://github.com/advisories/GHSA-8wr5-jm2h-8r4f)
- 2026-07-06: CVE-2026-54234: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. (https://nvd.nist.gov/vuln/detail/CVE-2026-54234)

## Sources

- 2026-07-06 NVD: CVE-2026-54234: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. https://nvd.nist.gov/vuln/detail/CVE-2026-54234
- 2026-07-17 GitHub Advisory Database: CVE-2026-54234: vLLM has Remote DoS via Invalid Recovered Token Reinjection https://github.com/advisories/GHSA-8wr5-jm2h-8r4f

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.

Related

  1. closed7.1
    vLLM vLLM denial of service via prompt embeds on M-RoPE models

    Patch now · Disclosed · since 20 Jul 2026 · updated 20 Jul 2026 · CVE-2026-55514

    AI
  2. closed8.7
    vLLM vLLM ReDoS vulnerability in structured outputs regex compilation

    Patch now · Disclosed · since 17 Jul 2026 · updated 17 Jul 2026 · CVE-2026-55574

    AI
  3. closed6.5
    vLLM CVE-2026-55646: vLLM speech-to-text enforces size limit after full file read

    Patch now · Disclosed · since 17 Jul 2026 · updated 17 Jul 2026 · CVE-2026-55646

    AI
  4. closed8.8
    vLLM CVE-2026-54232: Dependency confusion flaw in vLLM Dockerfile

    Patch now · Disclosed · since 22 Jun 2026 · CVE-2026-54232

    AI
  5. closed7.5
    vLLM CVE-2025-71379: vLLM regular expression denial of service vulnerabilities

    Patch now · Disclosed · since 20 Jun 2026 · CVE-2025-71379

    AI
  6. closed4.8
    vLLM CVE-2026-12491: vLLM flawed image metadata handling leads to input distortion

    Patch now · Disclosed · since 17 Jun 2026 · CVE-2026-12491

    AI