{"api_version":"1","generated_at":"2026-10-29T06:15:22+00:00","cve":"CVE-2026-115913","urls":{"html":"https://cve.report/CVE-2026-115913","api":"https://cve.report/api/cve/CVE-2026-115913.json","docs":"https://cve.report/api","cve_org":"https://www.cve.org/CVERecord?id=CVE-2026-115913","nvd":"https://nvd.nist.gov/vuln/detail/CVE-2026-115913"},"summary":{"title":"vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine","description":"vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.","state":"PUBLISHED","assigner":"GitHub_M","published_at":"2026-10-26 17:17:19","updated_at":"2026-10-28 15:38:12"},"problem_types":["CWE-770","CWE-770 CWE-770: Allocation of Resources Without Limits or Throttling"],"metrics":[{"version":"3.1","source":"security-advisories@github.com","type":"Secondary","score":"6.5","severity":"MEDIUM","vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","data":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","baseScore":6.5,"baseSeverity":"MEDIUM","attackVector":"NETWORK","attackComplexity":"LOW","privilegesRequired":"LOW","userInteraction":"NONE","scope":"UNCHANGED","confidentialityImpact":"NONE","integrityImpact":"NONE","availabilityImpact":"HIGH"}},{"version":"3.1","source":"CNA","type":"DECLARED","score":"6.5","severity":"MEDIUM","vector":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","data":{"attackComplexity":"LOW","attackVector":"NETWORK","availabilityImpact":"HIGH","baseScore":6.5,"baseSeverity":"MEDIUM","confidentialityImpact":"NONE","integrityImpact":"NONE","privilegesRequired":"LOW","scope":"UNCHANGED","userInteraction":"NONE","vectorString":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","version":"3.1"}}],"references":[{"url":"https://github.com/vllm-project/vllm/releases/tag/v0.32.0","name":"https://github.com/vllm-project/vllm/releases/tag/v0.32.0","refsource":"security-advisories@github.com","tags":["Release Notes"],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm","name":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm","refsource":"security-advisories@github.com","tags":["Exploit","Patch","Vendor Advisory"],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928","name":"https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928","refsource":"security-advisories@github.com","tags":["Patch"],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://github.com/vllm-project/vllm/pull/60614","name":"https://github.com/vllm-project/vllm/pull/60614","refsource":"security-advisories@github.com","tags":["Patch"],"title":"","mime":"","httpstatus":"","archivestatus":"0"},{"url":"https://www.cve.org/CVERecord?id=CVE-2026-115913","name":"CVE Program record","refsource":"CVE.ORG","tags":["canonical"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-115913","name":"NVD vulnerability detail","refsource":"NVD","tags":["canonical","analysis"]}],"affected":[{"source":"CNA","vendor":"vllm-project","product":"vllm","version":"affected < 0.32.0","platforms":[]}],"timeline":[],"solutions":[],"workarounds":[],"exploits":[],"credits":[],"nvd_cpes":[{"cve_year":"2026","cve_id":"115913","vulnerable":"1","versionEndIncluding":"","cpe1":"cpe","cpe2":"2.3","cpe3":"a","cpe4":"vllm","cpe5":"vllm","cpe6":"*","cpe7":"*","cpe8":"*","cpe9":"*","cpe10":"*","cpe11":"*","cpe12":"*","cpe13":"*"}],"vendor_comments":[],"enrichments":{"kev":null,"epss":{"cve_year":"2026","cve_id":"115913","cve":"CVE-2026-115913","epss":"0.003050000","percentile":"0.232050000","score_date":"2026-10-28","updated_at":"2026-10-29 00:09:14"},"legacy_qids":[]},"source_records":{"cve_program":{"containers":{"cna":{"affected":[{"product":"vllm","vendor":"vllm-project","versions":[{"status":"affected","version":"< 0.32.0"}]}],"descriptions":[{"lang":"en","value":"vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0."}],"metrics":[{"cvssV3_1":{"attackComplexity":"LOW","attackVector":"NETWORK","availabilityImpact":"HIGH","baseScore":6.5,"baseSeverity":"MEDIUM","confidentialityImpact":"NONE","integrityImpact":"NONE","privilegesRequired":"LOW","scope":"UNCHANGED","userInteraction":"NONE","vectorString":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","version":"3.1"}}],"problemTypes":[{"descriptions":[{"cweId":"CWE-770","description":"CWE-770: Allocation of Resources Without Limits or Throttling","lang":"en","type":"CWE"}]}],"providerMetadata":{"dateUpdated":"2026-10-26T16:56:59.817Z","orgId":"a0819718-46f1-4df5-94e2-005712e83aaa","shortName":"GitHub_M"},"references":[{"name":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm","tags":["x_refsource_CONFIRM"],"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm"},{"name":"https://github.com/vllm-project/vllm/pull/60614","tags":["x_refsource_MISC"],"url":"https://github.com/vllm-project/vllm/pull/60614"},{"name":"https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928","tags":["x_refsource_MISC"],"url":"https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928"},{"name":"https://github.com/vllm-project/vllm/releases/tag/v0.32.0","tags":["x_refsource_MISC"],"url":"https://github.com/vllm-project/vllm/releases/tag/v0.32.0"}],"source":{"advisory":"GHSA-8397-8qf6-f2gm","discovery":"UNKNOWN"},"title":"vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine"}},"cveMetadata":{"assignerOrgId":"a0819718-46f1-4df5-94e2-005712e83aaa","assignerShortName":"GitHub_M","cveId":"CVE-2026-115913","datePublished":"2026-10-26T16:56:59.817Z","dateReserved":"2026-10-23T20:21:45.107Z","dateUpdated":"2026-10-26T16:56:59.817Z","state":"PUBLISHED"},"dataType":"CVE_RECORD","dataVersion":"5.2"},"nvd":{"publishedDate":"2026-10-26 17:17:19","lastModifiedDate":"2026-10-28 15:38:12","problem_types":["CWE-770","CWE-770 CWE-770: Allocation of Resources Without Limits or Throttling"],"metrics":{"cvssMetricV31":[{"source":"security-advisories@github.com","type":"Secondary","cvssData":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H","baseScore":6.5,"baseSeverity":"MEDIUM","attackVector":"NETWORK","attackComplexity":"LOW","privilegesRequired":"LOW","userInteraction":"NONE","scope":"UNCHANGED","confidentialityImpact":"NONE","integrityImpact":"NONE","availabilityImpact":"HIGH"},"exploitabilityScore":2.8,"impactScore":3.6}]},"configurations":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*","versionStartIncluding":"0.31.0","versionEndExcluding":"0.32.0","matchCriteriaId":"50F36C60-8E0D-429B-9425-BFA2A4F71448"}]}]}]},"legacy_mitre":{"record":{"CveYear":"2026","CveId":"115913","Ordinal":"1","Title":"vLLM: Phi-4 multimodal audio encoder materializes and retains fu","CVE":"CVE-2026-115913","Year":"2026"},"notes":[{"CveYear":"2026","CveId":"115913","Ordinal":"1","NoteData":"vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.","Type":"Description","Title":"vLLM: Phi-4 multimodal audio encoder materializes and retains fu"}]}}}