{"id":"CVE-2026-100654","aliases":["EUVD-2026-87748","GHSA-9fp6-6v59-w2jx","GHSA-v5gm-qgmv-gc6c"],"status":"published","title":"vLLM before 0.29.0 Denial of Service via out-of-range stop_token_ids","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","severity":{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N","level":"high","assessor":"cna","from":"cve"},"epss":{"day":"2026-11-11","score":0.00314,"percentile":0.24579,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":false,"fix_available":false},"cwes":["CWE-129"],"published":"2026-09-26T13:23:23Z","modified":"2026-10-28T02:30:34Z","products":[{"vendor":"vllm","product":"vllm"},{"vendor":"vllm-project","product":"vllm"}],"references":[{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-v5gm-qgmv-gc6c","tags":["vendor-advisory","Mitigation","Vendor Advisory","advisory"]},{"url":"https://www.vulncheck.com/advisories/vllm-before-0.29.0-denial-of-service-via-out-of-range-stop-token-ids","tags":["third-party-advisory","Third Party Advisory","advisory"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-100654","tags":["advisory"]},{"url":"https://github.com/advisories/GHSA-9fp6-6v59-w2jx"},{"url":"https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/100xxx/CVE-2026-100654.json","tags":["advisory"]}],"sources":{"cve":[{"id":"CVE-2026-100654","title":"vLLM before 0.29.0 Denial of Service via out-of-range stop_token_ids","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","status":"published","published":"2026-09-26T13:23:23.312Z","modified":"2026-09-30T14:51:39.132Z","cvss":[{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N","level":"high","assessor":"cna"},{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium","assessor":"cna"}],"cwes":["CWE-129"],"updated_at":"2026-11-02T14:20:06.925367Z"}],"euvd":[{"id":"EUVD-2026-87748","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","status":"published","published":"2026-09-26T13:23:23Z","modified":"2026-09-30T14:51:39Z","cvss":[{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N","level":"high"}],"updated_at":"2026-11-02T17:27:49.499669Z"}],"ghsa":[{"id":"GHSA-9fp6-6v59-w2jx","title":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1...","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","status":"published","published":"2026-09-26T15:31:18Z","modified":"2026-09-26T15:31:23Z","cvss":[{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium"},{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","level":"high"}],"cwes":["CWE-129"],"updated_at":"2026-11-02T15:12:30.047516Z"}],"nvd":[{"id":"CVE-2026-100654","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","status":"published","published":"2026-09-26T14:16:48.1Z","modified":"2026-10-07T17:33:53.093Z","cvss":[{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","level":"high","assessor":"secondary"},{"version":"3.1","score":6.5,"vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","level":"medium","assessor":"secondary"}],"cwes":["CWE-129"],"updated_at":"2026-11-02T14:12:52.081323Z"}],"osv":[{"id":"CVE-2026-100654","title":"vLLM before 0.29.0 Denial of Service via out-of-range stop_token_ids","description":"vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens \u003e 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).","status":"published","published":"2026-09-26T13:23:23.312Z","modified":"2026-10-28T02:30:34.993573658Z","cvss":[{"version":"4.0","score":7.1,"vector":"CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N","level":"high"}],"cwes":["CWE-129"],"updated_at":"2026-11-02T15:44:53.490346Z"}]}}