{
 "id": "AIR-2026-0808",
 "slug": "cve-2026-54234-vllm-remote-denial-of-service-via-invalid-rec",
 "title": "CVE-2026-54234: vLLM remote denial of service via invalid recovered token reinjection",
 "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.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "vLLM"
 ],
 "incident_date": "2026-07-17",
 "incident_date_precision": "day",
 "first_reported": "2026-07-06",
 "sources": [
  {
   "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-54234",
   "publisher": "NVD",
   "title": "CVE-2026-54234: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.",
   "published_at": "2026-07-06"
  },
  {
   "url": "https://github.com/advisories/GHSA-8wr5-jm2h-8r4f",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2026-54234: vLLM has Remote DoS via Invalid Recovered Token Reinjection",
   "published_at": "2026-07-17"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2026-54234",
  "vllm",
  "denial-of-service"
 ],
 "created_at": "2026-10-18T07:00:14.101Z",
 "updated_at": "2026-10-18T10:04:21.885Z"
}