{"api_version":"1","generated_at":"2026-07-07T18:18:26+00:00","cve":"CVE-2026-54234","urls":{"html":"https://cve.report/CVE-2026-54234","api":"https://cve.report/api/cve/CVE-2026-54234.json","docs":"https://cve.report/api","cve_org":"https://www.cve.org/CVERecord?id=CVE-2026-54234","nvd":"https://nvd.nist.gov/vuln/detail/CVE-2026-54234"},"summary":{"title":"vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection","description":"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.","state":"PUBLISHED","assigner":"GitHub_M","published_at":"2026-07-06 21:16:56","updated_at":"2026-07-07 15:16:48"},"problem_types":["CWE-20","CWE-1284","CWE-20 CWE-20: Improper Input Validation","CWE-1284 CWE-1284: Improper Validation of Specified Quantity in Input"],"metrics":[{"version":"3.1","source":"security-advisories@github.com","type":"Secondary","score":"7.5","severity":"HIGH","vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","data":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","baseScore":7.5,"baseSeverity":"HIGH","attackVector":"NETWORK","attackComplexity":"LOW","privilegesRequired":"NONE","userInteraction":"NONE","scope":"UNCHANGED","confidentialityImpact":"NONE","integrityImpact":"NONE","availabilityImpact":"HIGH"}},{"version":"3.1","source":"CNA","type":"DECLARED","score":"7.5","severity":"HIGH","vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","data":{"attackComplexity":"LOW","attackVector":"NETWORK","availabilityImpact":"HIGH","baseScore":7.5,"baseSeverity":"HIGH","confidentialityImpact":"NONE","integrityImpact":"NONE","privilegesRequired":"NONE","scope":"UNCHANGED","userInteraction":"NONE","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","version":"3.1"}}],"references":[{"url":"https://github.com/vllm-project/vllm/pull/44744","name":"https://github.com/vllm-project/vllm/pull/44744","refsource":"security-advisories@github.com","tags":[],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca","name":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca","refsource":"security-advisories@github.com","tags":[],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f","name":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f","refsource":"134c704f-9b21-4f2e-91b3-4a467353bcc0","tags":[],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://www.cve.org/CVERecord?id=CVE-2026-54234","name":"CVE Program record","refsource":"CVE.ORG","tags":["canonical"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-54234","name":"NVD vulnerability detail","refsource":"NVD","tags":["canonical","analysis"]}],"affected":[{"source":"CNA","vendor":"vllm-project","product":"vllm","version":"affected < 0.24.0","platforms":[]}],"timeline":[],"solutions":[],"workarounds":[],"exploits":[],"credits":[],"nvd_cpes":[],"vendor_comments":[],"enrichments":{"kev":null,"epss":null,"legacy_qids":[]},"source_records":{"cve_program":{"containers":{"adp":[{"metrics":[{"other":{"content":{"id":"CVE-2026-54234","options":[{"Exploitation":"poc"},{"Automatable":"yes"},{"Technical Impact":"partial"}],"role":"CISA Coordinator","timestamp":"2026-07-07T14:12:43.565166Z","version":"2.0.3"},"type":"ssvc"}}],"providerMetadata":{"dateUpdated":"2026-07-07T14:13:34.348Z","orgId":"134c704f-9b21-4f2e-91b3-4a467353bcc0","shortName":"CISA-ADP"},"references":[{"tags":["exploit"],"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f"}],"title":"CISA ADP Vulnrichment"}],"cna":{"affected":[{"product":"vllm","vendor":"vllm-project","versions":[{"status":"affected","version":"< 0.24.0"}]}],"descriptions":[{"lang":"en","value":"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."}],"metrics":[{"cvssV3_1":{"attackComplexity":"LOW","attackVector":"NETWORK","availabilityImpact":"HIGH","baseScore":7.5,"baseSeverity":"HIGH","confidentialityImpact":"NONE","integrityImpact":"NONE","privilegesRequired":"NONE","scope":"UNCHANGED","userInteraction":"NONE","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","version":"3.1"}}],"problemTypes":[{"descriptions":[{"cweId":"CWE-20","description":"CWE-20: Improper Input Validation","lang":"en","type":"CWE"}]},{"descriptions":[{"cweId":"CWE-1284","description":"CWE-1284: Improper Validation of Specified Quantity in Input","lang":"en","type":"CWE"}]}],"providerMetadata":{"dateUpdated":"2026-07-06T19:49:20.481Z","orgId":"a0819718-46f1-4df5-94e2-005712e83aaa","shortName":"GitHub_M"},"references":[{"name":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f","tags":["x_refsource_CONFIRM"],"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f"},{"name":"https://github.com/vllm-project/vllm/pull/44744","tags":["x_refsource_MISC"],"url":"https://github.com/vllm-project/vllm/pull/44744"},{"name":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca","tags":["x_refsource_MISC"],"url":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca"}],"source":{"advisory":"GHSA-8wr5-jm2h-8r4f","discovery":"UNKNOWN"},"title":"vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection"}},"cveMetadata":{"assignerOrgId":"a0819718-46f1-4df5-94e2-005712e83aaa","assignerShortName":"GitHub_M","cveId":"CVE-2026-54234","datePublished":"2026-07-06T19:49:20.481Z","dateReserved":"2026-06-12T16:25:43.084Z","dateUpdated":"2026-07-07T14:13:34.348Z","state":"PUBLISHED"},"dataType":"CVE_RECORD","dataVersion":"5.2"},"nvd":{"publishedDate":"2026-07-06 21:16:56","lastModifiedDate":"2026-07-07 15:16:48","problem_types":["CWE-20","CWE-1284","CWE-20 CWE-20: Improper Input Validation","CWE-1284 CWE-1284: Improper Validation of Specified Quantity in Input"],"metrics":{"cvssMetricV31":[{"source":"security-advisories@github.com","type":"Secondary","cvssData":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","baseScore":7.5,"baseSeverity":"HIGH","attackVector":"NETWORK","attackComplexity":"LOW","privilegesRequired":"NONE","userInteraction":"NONE","scope":"UNCHANGED","confidentialityImpact":"NONE","integrityImpact":"NONE","availabilityImpact":"HIGH"},"exploitabilityScore":3.9,"impactScore":3.6}],"ssvcV203":[{"source":"134c704f-9b21-4f2e-91b3-4a467353bcc0","ssvcData":{"timestamp":"2026-07-07T14:12:43.565166Z","id":"CVE-2026-54234","options":[{"exploitation":"poc"},{"automatable":"yes"},{"technicalImpact":"partial"}],"role":"CISA Coordinator","version":"2.0.3"}}]},"configurations":[]},"legacy_mitre":{"record":{"CveYear":"2026","CveId":"54234","Ordinal":"1","Title":"vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection","CVE":"CVE-2026-54234","Year":"2026"},"notes":[{"CveYear":"2026","CveId":"54234","Ordinal":"1","NoteData":"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.","Type":"Description","Title":"vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection"}]}}}