{
 "id": "AIR-2025-0521",
 "slug": "cve-2025-62164-vllm-memory-corruption-via-unvalidated-prompt",
 "title": "CVE-2025-62164: vLLM memory corruption via unvalidated prompt-embedding tensors",
 "summary": "vLLM from 0.10.2 before 0.11.1 loaded user-supplied prompt embeddings in its Completions API with torch.load() and no sparse tensor validation, so a crafted tensor could write out of bounds during to_dense(). The flaw crashes the server and may allow remote code execution; 0.11.1 fixes it.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "vLLM"
 ],
 "incident_date": "2025-11-20",
 "incident_date_precision": "day",
 "first_reported": "2025-11-20",
 "sources": [
  {
   "url": "https://github.com/advisories/GHSA-mrw7-hf4f-83pf",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2025-62164: vLLM deserialization vulnerability leading to DoS and potential RCE",
   "published_at": "2025-11-20"
  },
  {
   "url": "https://nvd.nist.gov/vuln/detail/CVE-2025-62164",
   "publisher": "NVD",
   "title": "CVE-2025-62164: vLLM is an inference and serving engine for large language models (LLMs).",
   "published_at": "2025-11-21"
  }
 ],
 "source_count": 2,
 "confidence": 0.9,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2025-62164",
  "vllm",
  "memory-corruption",
  "dos"
 ],
 "created_at": "2026-10-18T18:20:34.882Z",
 "updated_at": "2026-10-18T18:20:34.882Z"
}