{"id":"CVE-2026-54234","state":"PUBLISHED","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.","publishedAt":"2026-07-06T19:49:20.481Z","updatedAt":"2026-07-07T14:13:34.348Z","severity":{"score":7.5,"label":"High","version":"3.1","vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","source":"GitHub_M"},"assigner":"GitHub_M","products":["vllm-project vllm"],"affected":[{"vendor":"vllm-project","product":"vllm","versions":[{"version":"< 0.24.0","status":"affected"}]}],"references":["https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f","https://github.com/vllm-project/vllm/pull/44744","https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca"],"weaknesses":["CWE-20","CWE-1284"],"ingestedAt":"2026-10-29T20:12:24.861Z","source":{"repository":"CVEProject/cvelistV5","path":"cves/2026/54xxx/CVE-2026-54234.json","commit":"556c9d3a8cd17e218bb8e452ba7cd619a384ff47"},"raw":{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-54234","state":"PUBLISHED","datePublished":"2026-07-06T19:49:20.481Z","dateUpdated":"2026-07-07T14:13:34.348Z","assignerShortName":"GitHub_M","assignerOrgId":"a0819718-46f1-4df5-94e2-005712e83aaa","dateReserved":"2026-06-12T16:25:43.084Z"},"containers":{"cna":{"title":"vLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection","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."}],"affected":[{"vendor":"vllm-project","product":"vllm","versions":[{"version":"< 0.24.0","status":"affected"}]}],"references":[{"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","tags":["x_refsource_CONFIRM"]},{"url":"https://github.com/vllm-project/vllm/pull/44744","name":"https://github.com/vllm-project/vllm/pull/44744","tags":["x_refsource_MISC"]},{"url":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca","name":"https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca","tags":["x_refsource_MISC"]}],"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","lang":"en","description":"CWE-20: Improper Input Validation","type":"CWE"}]},{"descriptions":[{"cweId":"CWE-1284","lang":"en","description":"CWE-1284: Improper Validation of Specified Quantity in Input","type":"CWE"}]}],"providerMetadata":{"shortName":"GitHub_M","orgId":"a0819718-46f1-4df5-94e2-005712e83aaa","dateUpdated":"2026-07-06T19:49:20.481Z"},"source":{"advisory":"GHSA-8wr5-jm2h-8r4f","discovery":"UNKNOWN"}},"adp":[{"title":"CISA ADP Vulnrichment","references":[{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f","tags":["exploit"]}],"metrics":[{"other":{"type":"ssvc","content":{"timestamp":"2026-07-07T14:12:43.565166Z","id":"CVE-2026-54234","options":[{"Exploitation":"poc"},{"Automatable":"yes"},{"Technical Impact":"partial"}],"role":"CISA Coordinator","version":"2.0.3"}}}],"providerMetadata":{"shortName":"CISA-ADP","orgId":"134c704f-9b21-4f2e-91b3-4a467353bcc0","dateUpdated":"2026-07-07T14:13:34.348Z"}}]}}}