{
 "id": "AIR-2025-0353",
 "slug": "cve-2025-61620-vllm-resource-exhaustion-via-malicious-jinja",
 "title": "CVE-2025-61620: vLLM resource exhaustion via malicious Jinja template",
 "summary": "A denial-of-service vulnerability exists in vLLM's OpenAI-compatible server when handling user-specified Jinja templates. Attackers can exhaust CPU and memory resources to cause a service outage.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "vLLM"
 ],
 "incident_date": "2025-10-07",
 "incident_date_precision": "day",
 "first_reported": "2025-10-07",
 "sources": [
  {
   "url": "https://github.com/advisories/GHSA-6fvq-23cw-5628",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2025-61620: vLLM: Resource-Exhaustion (DoS) through Malicious Jinja Template in OpenAI-Compatible Server",
   "published_at": "2025-10-07"
  }
 ],
 "source_count": 1,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2025-61620",
  "dos",
  "vllm"
 ],
 "created_at": "2026-10-18T07:00:14.101Z",
 "updated_at": "2026-10-18T07:00:14.101Z"
}