{
 "id": "AIR-2025-0278",
 "slug": "cve-2025-46560-vllm-vulnerable-to-denial-of-service-via-quad",
 "title": "CVE-2025-46560: vLLM vulnerable to denial of service via quadratic time complexity",
 "summary": "A vulnerability in vLLM's multimodal tokenizer input processor causes quadratic time complexity leading to denial of service. The issue allows malicious actors to trigger resource exhaustion via crafted inputs.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "vLLM"
 ],
 "incident_date": "2025-04-29",
 "incident_date_precision": "day",
 "first_reported": "2025-04-29",
 "sources": [
  {
   "url": "https://github.com/advisories/GHSA-vc6m-hm49-g9qg",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2025-46560: vLLM: Quadratic Time Complexity in Input Token Processing​ leads to denial of service",
   "published_at": "2025-04-29"
  },
  {
   "url": "https://nvd.nist.gov/vuln/detail/CVE-2025-46560",
   "publisher": "NVD",
   "title": "CVE-2025-46560: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.",
   "published_at": "2025-04-30"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2025-46560",
  "dos",
  "vllm"
 ],
 "created_at": "2026-10-18T07:00:14.101Z",
 "updated_at": "2026-10-18T10:04:21.885Z"
}