{
  "schemaVersion": 1,
  "attribution": [
    {
      "source": "CISA Known Exploited Vulnerabilities",
      "url": "https://www.cisa.gov/known-exploited-vulnerabilities-catalog",
      "licence": "US government work"
    },
    {
      "source": "GitHub Advisory Database",
      "url": "https://github.com/advisories",
      "licence": "CC BY 4.0"
    },
    {
      "source": "Have I Been Pwned",
      "url": "https://haveibeenpwned.com/",
      "licence": "CC BY 4.0"
    },
    {
      "source": "CVE Program (CVE records)",
      "url": "https://www.cve.org/Legal/TermsOfUse",
      "licence": "CVE Terms of Use",
      "notice": "Copyright © 1999-2026, The MITRE Corporation. CVE is a trademark and the CVE logo is a registered trademark of The MITRE Corporation.",
      "terms": "CVE Usage: MITRE hereby grants you a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare derivative works of, publicly display, publicly perform, sublicense, and distribute Common Vulnerabilities and Exposures (CVE™). Any copy you make for such purposes is authorized provided that you reproduce MITRE's copyright designation and this license in any such copy."
    },
    {
      "source": "FIRST EPSS",
      "url": "https://www.first.org/epss/",
      "licence": "free to use; FIRST requests attribution"
    }
  ],
  "record": {
    "id": "CVE-2026-54234",
    "sourceURL": "https://www.cve.org/CVERecord?id=CVE-2026-54234",
    "idYear": 2026,
    "status": "PUBLISHED",
    "publishedAt": "2026-07-06T19:49:20.481Z",
    "updatedAt": "2026-07-07T14:13:34.348Z",
    "importedAt": "2026-10-24T01:58:20.586Z",
    "snapshot": {
      "id": "cve_2026-10-24_midnight_baseline",
      "at": "2026-10-24T00:00:00.000Z",
      "source": "https://github.com/CVEProject/cvelistV5/releases/tag/cve_2026-10-24_0100Z"
    },
    "facts": {
      "title": "vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection",
      "summary": "vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, a frontend-legal multi-request speculative decoding workload can cause the rejection sampler to produce a recovered token equal to the model vocabulary size boundary value, which is then converted to negative one when the engine selects the next live token for a request and is written back into the drafter's input ids; that out-of-vocabulary value is later consumed by the model's embedding and attention path and crashes the engine worker with a GPU device-side assertion. The same triggering request sequence is reachable through the public gRPC Generate and Abort endpoints, so a remote client that can send generation requests can crash the shared engine worker, aborting concurrent requests and causing a service-wide denial of service for other clients of the deployment until the worker is restarted. This issue is fixed in version 0.24.0.",
      "affected": [
        {
          "vendor": "vllm-project",
          "product": "vllm",
          "versions": [
            {
              "version": "< 0.24.0",
              "status": "affected"
            }
          ]
        }
      ]
    }
  }
}