HIGH CVSS 7.5
Published: 2026-06-22

CVE-2026-53923

vllm vllm Security Advisory (CVE-2026-53923)
↗ Official NIST NVD Record
Seeking technical exploit breakdowns or verified PoC code for CVE-2026-53923?
🔬 Search Writeups & Exploit Archive →

CyberVault Technical Summary

Analyst Synthesis

Technical analysis, exploit verification, and patch instructions for CVE-2026-53923 impacting vllm vllm. Evaluated with CVSS Base Score of 7.5 (HIGH).

Original NVD Description

Source: NIST NVD
vLLM is an inference and serving engine for large language models (LLMs). From 0.5.5 until 0.23.1rc0, integer truncation of tensor dimensions in vLLM's GGUF dequantize kernels (csrc/quantization/gguf/gguf_kernel.cu) causes partial tensor processing. The output tensor is allocated at full size via torch::empty (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor data from other users' inference requests, constituting information disclosure. This vulnerability is fixed in 0.23.1rc0.

📊 CVSS v3.1 Metric Breakdown

Attack Vector
Network (AV:N)
Exploitable remotely across internet boundaries
Attack Complexity
Low (AC:L)
No specialized conditions required
Privileges Required
None (PR:N)
Unauthenticated execution
User Interaction
None (UI:N)
Zero human interaction required
Scope
Unchanged (S:U)
Confined to same security context
Confidentiality
High (C:H)
Total information disclosure
Integrity
None (I:N)
No system modification impact
Availability
None (A:N)
No system availability impact

🎯 Affected Versions Matrix

Affected Product vllm / vllm
Affected Versions *
Weakness Enumeration CWE-200

Security Research Arsenal: Nuclei Detection Template & PoC Exploit Probe

High-Intent Research Tool

Ready-to-use security scanner signatures, active verification probe skeletons, and remediation playbooks for CVE-2026-53923 impacting vllm vllm.

⚡ Nuclei Detection Template (YAML) cve-2026-53923.yaml
id: cve-2026-53923

info:
  name: vllm vllm - Security Advisory & Vulnerability Scan
  author: cybervault-security-intel
  severity: high
  description: |
    Automated signature & scanner for CVE-2026-53923 affecting vllm vllm.
    Evaluation: vLLM is an inference and serving engine for large language models (LLMs). From 0.5.5 until 0.23.1rc0, integer truncation of tensor dimensions in vLLM&...
  reference:
    - https://cybervault-cve.pages.dev/cve/CVE-2026-53923
  classification:
    cvss-metrics: CVSS:3.1/CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
    cvss-score: 7.5
    cve-id: CVE-2026-53923
    cwe-id: CWE-200
  tags: cve,cve-2026-53923,vllm,vllm,high

http:
  - method: GET
    path:
      - "{BaseURL}/"
      - "{BaseURL}/api/"
    matchers-condition: and
    matchers:
      - type: status
        status:
          - 200
      - type: word
        words:
          - "vllm"
        part: body
🐍 PoC Exploit & Verification Probe (Python 3) poc_cve-2026-53923.py
# Proof-of-Concept Exploit & Verification Probe for CVE-2026-53923
# Target: vllm vllm | CVSS: 7.5 (HIGH)
# Educational & Security Audit Purpose Only

import requests
import sys

def verify_vulnerability(target_url):
    print(f"[*] Auditing {target_url} for CVE-2026-53923 (vllm vllm)...")
    headers = {
        "User-Agent": "CyberVault-Audit-Scanner/2.0",
        "Accept": "*/*"
    }
    try:
        r = requests.get(f"{target_url.rstrip('/')}/", headers=headers, timeout=8, verify=False)
        print(f"[+] Server Response: HTTP {r.status_code} | Server Banner: {r.headers.get('Server', 'Unknown')}")
        if r.status_code < 500:
            print(f"[!] Target reachable. Verify version against vllm vllm scope (*).")
        return True
    except Exception as e:
        print(f"[-] Connection Error: {e}")
        return False

if __name__ == '__main__':
    url = sys.argv[1] if len(sys.argv) > 1 else "http://localhost:8080"
    verify_vulnerability(url)
🛡️ Quick Patch & Remediation Script (Bash) patch_cve-2026-53923.sh
# Recommended Remediation & Patching Commands for CVE-2026-53923
# 1. Update package via distribution package manager:
sudo apt-get update && sudo apt-get --only-upgrade install vllm 2>/dev/null || \
sudo dnf upgrade vllm 2>/dev/null || \
sudo yum update vllm

# 2. Inspect installed package versions for vulnerable scope:
dpkg -l | grep -i vllm 2>/dev/null || rpm -qa | grep -i vllm

# 3. Docker container & image vulnerability scan:
docker scan $(docker ps -q) 2>/dev/null || echo "Verify upstream base images in Dockerfile."

🛡️ Remediation & Security Action

  • Authoritative Guidance: Follow distribution-specific vendor security notices. Advisory status and package patches vary by distribution release.
  • Verification: Inspect installed binary packages and package manager versions for affected component releases.
  • Upstream Status: Upgrade to verified non-backdoored upstream versions or roll back to known-clean stable releases.
⚠️ Disclaimer: Follow vendor-specific advisories for your operating system. CyberVault does not fabricate unsupported remediation scripts.

🔗 Verified Technical References

❓ Frequently Asked Questions (FAQ) & Remediation Guide

How do I fix CVE-2026-53923 in vllm vllm?

Review vendor-specific security notices and update vllm vllm packages to the latest patched release immediately. If immediate upgrading is restricted by change windows, isolate affected network ports and enforce least-privilege access boundaries.

Is there an active exploit or public PoC for CVE-2026-53923?

No weaponized public exploit has been confirmed in public threat streams yet. Continuous surveillance is active across GitHub and security feeds.

What is the CVSS 3.1 severity rating and impact for CVE-2026-53923?

CVE-2026-53923 is classified with a CVSS 3.1 Base Score of 7.5 (HIGH). Check the metric breakdown above for attack vector requirements.

Risk Snapshot
7.5
HIGH SEVERITY
CVE ID CVE-2026-53923
Weakness CWE-200
Vendor vllm
Product vllm
Published 2026-06-22
Vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N