Vulnerability-Lookup

CVE-2026-105760 (GCVE-0-2026-105760)

Vulnerability from cvelistv5 – Published: 2026-10-05 23:01 – Updated: 2026-10-06 14:39

Title

vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion

Summary

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0.

Severity

5.3 (Medium)


                        
                          CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L

SSVC

Exploitation: none Automatable: yes Technical Impact: partial

CISA Coordinator · CISA-ADP (v2.0.3)

Decision recorded 2026-10-06 14:38 UTC

CWE

CWE-400 - Uncontrolled Resource Consumption

Assigner

GitHub_M CNA under the mitre root

CNA scorecard C 72/100 over 8010 records in the last 180 days details

References

4 references

URL	Tags
https://github.com/vllm-project/vllm/security/adv…	x_refsource_CONFIRM
https://github.com/vllm-project/vllm/pull/54935	x_refsource_MISC
https://github.com/vllm-project/vllm/commit/8b6de…	x_refsource_MISC
https://github.com/vllm-project/vllm/releases/tag…	x_refsource_MISC

Impacted products

1 product

Vendor	Product	Version	CPE status
vllm-project	vllm	Affected: >= 0.23.0rc2, < 0.30.0	guessed

Show details on NVD website

JSON

To clipboard

{
  "containers": {
    "adp": [
      {
        "metrics": [
          {
            "other": {
              "content": {
                "id": "CVE-2026-105760",
                "options": [
                  {
                    "Exploitation": "none"
                  },
                  {
                    "Automatable": "yes"
                  },
                  {
                    "Technical Impact": "partial"
                  }
                ],
                "role": "CISA Coordinator",
                "timestamp": "2026-10-06T14:38:28.658453Z",
                "version": "2.0.3"
              },
              "type": "ssvc"
            }
          }
        ],
        "providerMetadata": {
          "dateUpdated": "2026-10-06T14:39:17.058Z",
          "orgId": "134c704f-9b21-4f2e-91b3-4a467353bcc0",
          "shortName": "CISA-ADP"
        },
        "title": "CISA ADP Vulnrichment"
      }
    ],
    "cna": {
      "affected": [
        {
          "product": "vllm",
          "vendor": "vllm-project",
          "versions": [
            {
              "status": "affected",
              "version": "\u003e= 0.23.0rc2, \u003c 0.30.0"
            }
          ]
        }
      ],
      "descriptions": [
        {
          "lang": "en",
          "value": "vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0."
        }
      ],
      "metrics": [
        {
          "cvssV3_1": {
            "attackComplexity": "LOW",
            "attackVector": "NETWORK",
            "availabilityImpact": "LOW",
            "baseScore": 5.3,
            "baseSeverity": "MEDIUM",
            "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:L",
            "version": "3.1"
          }
        }
      ],
      "problemTypes": [
        {
          "descriptions": [
            {
              "cweId": "CWE-400",
              "description": "CWE-400: Uncontrolled Resource Consumption",
              "lang": "en",
              "type": "CWE"
            }
          ]
        }
      ],
      "providerMetadata": {
        "dateUpdated": "2026-10-05T23:01:54.972Z",
        "orgId": "a0819718-46f1-4df5-94e2-005712e83aaa",
        "shortName": "GitHub_M"
      },
      "references": [
        {
          "name": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr",
          "tags": [
            "x_refsource_CONFIRM"
          ],
          "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
        },
        {
          "name": "https://github.com/vllm-project/vllm/pull/54935",
          "tags": [
            "x_refsource_MISC"
          ],
          "url": "https://github.com/vllm-project/vllm/pull/54935"
        },
        {
          "name": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7",
          "tags": [
            "x_refsource_MISC"
          ],
          "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
        },
        {
          "name": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0",
          "tags": [
            "x_refsource_MISC"
          ],
          "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
        }
      ],
      "source": {
        "advisory": "GHSA-58v5-2m8f-94pr",
        "discovery": "UNKNOWN"
      },
      "title": "vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion"
    }
  },
  "cveMetadata": {
    "assignerOrgId": "a0819718-46f1-4df5-94e2-005712e83aaa",
    "assignerShortName": "GitHub_M",
    "cveId": "CVE-2026-105760",
    "datePublished": "2026-10-05T23:01:54.972Z",
    "dateReserved": "2026-10-05T19:11:07.948Z",
    "dateUpdated": "2026-10-06T14:39:17.058Z",
    "state": "PUBLISHED"
  },
  "dataType": "CVE_RECORD",
  "dataVersion": "5.2",
  "vulnerability-lookup:meta": {
    "epss": {
      "cve": "CVE-2026-105760",
      "date": "2026-10-30",
      "epss": "0.00297",
      "percentile": "0.22375"
    },
    "nvd": {
      "cve": {
        "affected": [
          {
            "affectedData": [
              {
                "product": "vllm",
                "vendor": "vllm-project",
                "versions": [
                  {
                    "status": "affected",
                    "version": "\u003e= 0.23.0rc2, \u003c 0.30.0"
                  }
                ]
              }
            ],
            "source": "security-advisories@github.com"
          }
        ],
        "configurations": [
          {
            "nodes": [
              {
                "cpeMatch": [
                  {
                    "criteria": "cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*",
                    "matchCriteriaId": "EF515F5D-C23A-4F8D-BF22-864E2A92BA0E",
                    "versionEndExcluding": "0.30.0",
                    "versionStartIncluding": "0.23.0",
                    "vulnerable": true
                  }
                ],
                "negate": false,
                "operator": "OR"
              }
            ]
          }
        ],
        "cveTags": [],
        "descriptions": [
          {
            "lang": "en",
            "value": "vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0."
          }
        ],
        "id": "CVE-2026-105760",
        "lastModified": "2026-10-16T20:39:54.097",
        "metrics": {
          "cvssMetricV31": [
            {
              "cvssData": {
                "attackComplexity": "LOW",
                "attackVector": "NETWORK",
                "availabilityImpact": "LOW",
                "baseScore": 5.3,
                "baseSeverity": "MEDIUM",
                "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:L",
                "version": "3.1"
              },
              "exploitabilityScore": 3.9,
              "impactScore": 1.4,
              "source": "security-advisories@github.com",
              "type": "Secondary"
            }
          ],
          "ssvcV203": [
            {
              "source": "134c704f-9b21-4f2e-91b3-4a467353bcc0",
              "ssvcData": {
                "id": "CVE-2026-105760",
                "options": [
                  {
                    "exploitation": "none"
                  },
                  {
                    "automatable": "yes"
                  },
                  {
                    "technicalImpact": "partial"
                  }
                ],
                "role": "CISA Coordinator",
                "timestamp": "2026-10-06T14:38:28.658453Z",
                "version": "2.0.3"
              }
            }
          ]
        },
        "published": "2026-10-05T23:17:02.903",
        "references": [
          {
            "source": "security-advisories@github.com",
            "tags": [
              "Patch"
            ],
            "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
          },
          {
            "source": "security-advisories@github.com",
            "tags": [
              "Issue Tracking",
              "Patch"
            ],
            "url": "https://github.com/vllm-project/vllm/pull/54935"
          },
          {
            "source": "security-advisories@github.com",
            "tags": [
              "Release Notes"
            ],
            "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
          },
          {
            "source": "security-advisories@github.com",
            "tags": [
              "Mitigation",
              "Vendor Advisory"
            ],
            "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
          }
        ],
        "sourceIdentifier": "security-advisories@github.com",
        "vulnStatus": "Analyzed",
        "weaknesses": [
          {
            "description": [
              {
                "lang": "en",
                "value": "CWE-400"
              }
            ],
            "source": "security-advisories@github.com",
            "type": "Secondary"
          }
        ]
      }
    },
    "redhat_vex": {
      "aggregate_severity": "Moderate",
      "current_release_date": "2026-10-15T23:33:35+00:00",
      "cve": "CVE-2026-105760",
      "id": "CVE-2026-105760",
      "initial_release_date": "2026-10-05T23:01:54.972000+00:00",
      "product_status:known_affected": "27",
      "source": "Red Hat CSAF VEX",
      "status": "final",
      "title": "vllm: vllm: Denial of Service via excessive resource consumption in GLMGA video backend",
      "url": "https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-105760.json",
      "version": "3"
    },
    "vulnrichment": {
      "containers": {
        "adp": [
          {
            "metrics": [
              {
                "other": {
                  "content": {
                    "id": "CVE-2026-105760",
                    "options": [
                      {
                        "Exploitation": "none"
                      },
                      {
                        "Automatable": "yes"
                      },
                      {
                        "Technical Impact": "partial"
                      }
                    ],
                    "role": "CISA Coordinator",
                    "timestamp": "2026-10-06T14:38:28.658453Z",
                    "version": "2.0.3"
                  },
                  "type": "ssvc"
                }
              }
            ],
            "providerMetadata": {
              "dateUpdated": "2026-10-06T14:39:09.803Z",
              "orgId": "134c704f-9b21-4f2e-91b3-4a467353bcc0",
              "shortName": "CISA-ADP"
            },
            "title": "CISA ADP Vulnrichment"
          }
        ],
        "cna": {
          "affected": [
            {
              "product": "vllm",
              "vendor": "vllm-project",
              "versions": [
                {
                  "status": "affected",
                  "version": "\u003e= 0.23.0rc2, \u003c 0.30.0"
                }
              ]
            }
          ],
          "descriptions": [
            {
              "lang": "en",
              "value": "vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0."
            }
          ],
          "metrics": [
            {
              "cvssV3_1": {
                "attackComplexity": "LOW",
                "attackVector": "NETWORK",
                "availabilityImpact": "LOW",
                "baseScore": 5.3,
                "baseSeverity": "MEDIUM",
                "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:L",
                "version": "3.1"
              }
            }
          ],
          "problemTypes": [
            {
              "descriptions": [
                {
                  "cweId": "CWE-400",
                  "description": "CWE-400: Uncontrolled Resource Consumption",
                  "lang": "en",
                  "type": "CWE"
                }
              ]
            }
          ],
          "providerMetadata": {
            "dateUpdated": "2026-10-05T23:01:54.972Z",
            "orgId": "a0819718-46f1-4df5-94e2-005712e83aaa",
            "shortName": "GitHub_M"
          },
          "references": [
            {
              "name": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr",
              "tags": [
                "x_refsource_CONFIRM"
              ],
              "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
            },
            {
              "name": "https://github.com/vllm-project/vllm/pull/54935",
              "tags": [
                "x_refsource_MISC"
              ],
              "url": "https://github.com/vllm-project/vllm/pull/54935"
            },
            {
              "name": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7",
              "tags": [
                "x_refsource_MISC"
              ],
              "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
            },
            {
              "name": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0",
              "tags": [
                "x_refsource_MISC"
              ],
              "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
            }
          ],
          "source": {
            "advisory": "GHSA-58v5-2m8f-94pr",
            "discovery": "UNKNOWN"
          },
          "title": "vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion"
        }
      },
      "cveMetadata": {
        "assignerOrgId": "a0819718-46f1-4df5-94e2-005712e83aaa",
        "assignerShortName": "GitHub_M",
        "cveId": "CVE-2026-105760",
        "datePublished": "2026-10-05T23:01:54.972Z",
        "dateReserved": "2026-10-05T19:11:07.948Z",
        "dateUpdated": "2026-10-05T23:01:54.972Z",
        "state": "PUBLISHED"
      },
      "dataType": "CVE_RECORD",
      "dataVersion": "5.2"
    }
  }
}

FKIE_CVE-2026-105760

Vulnerability from fkie_nvd - Published: 2026-10-05 23:17 - Updated: 2026-10-16 20:39

Severity

5.3 (Medium) - CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L

Summary

References

URL	Tags
security-advisories@github.com	https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7	Patch
security-advisories@github.com	https://github.com/vllm-project/vllm/pull/54935	Issue Tracking, Patch
security-advisories@github.com	https://github.com/vllm-project/vllm/releases/tag/v0.30.0	Release Notes
security-advisories@github.com	https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr	Mitigation, Vendor Advisory

Impacted products

	Vendor	Product	Version
	vllm	vllm	*

JSON

To clipboard

{
  "affected": [
    {
      "affectedData": [
        {
          "product": "vllm",
          "vendor": "vllm-project",
          "versions": [
            {
              "status": "affected",
              "version": "\u003e= 0.23.0rc2, \u003c 0.30.0"
            }
          ]
        }
      ],
      "source": "security-advisories@github.com"
    }
  ],
  "configurations": [
    {
      "nodes": [
        {
          "cpeMatch": [
            {
              "criteria": "cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*",
              "matchCriteriaId": "EF515F5D-C23A-4F8D-BF22-864E2A92BA0E",
              "versionEndExcluding": "0.30.0",
              "versionStartIncluding": "0.23.0",
              "vulnerable": true
            }
          ],
          "negate": false,
          "operator": "OR"
        }
      ]
    }
  ],
  "cveTags": [],
  "descriptions": [
    {
      "lang": "en",
      "value": "vLLM is an inference and serving engine for large language models. Prior to 0.30.0, a caller can use the request-level media_io_kwargs field to select the GLMGA video backend and supply large values for the fps and max_frames options without a strict work ceiling. GLMGA constructs and deduplicates an attacker-sized pre-decode frame-index list, allowing a compact request and tiny valid video to consume disproportionate CPU time and memory in the shared media-loading executor. This issue is fixed in version 0.30.0."
    }
  ],
  "id": "CVE-2026-105760",
  "lastModified": "2026-10-16T20:39:54.097",
  "metrics": {
    "cvssMetricV31": [
      {
        "cvssData": {
          "attackComplexity": "LOW",
          "attackVector": "NETWORK",
          "availabilityImpact": "LOW",
          "baseScore": 5.3,
          "baseSeverity": "MEDIUM",
          "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:L",
          "version": "3.1"
        },
        "exploitabilityScore": 3.9,
        "impactScore": 1.4,
        "source": "security-advisories@github.com",
        "type": "Secondary"
      }
    ],
    "ssvcV203": [
      {
        "source": "134c704f-9b21-4f2e-91b3-4a467353bcc0",
        "ssvcData": {
          "id": "CVE-2026-105760",
          "options": [
            {
              "exploitation": "none"
            },
            {
              "automatable": "yes"
            },
            {
              "technicalImpact": "partial"
            }
          ],
          "role": "CISA Coordinator",
          "timestamp": "2026-10-06T14:38:28.658453Z",
          "version": "2.0.3"
        }
      }
    ]
  },
  "published": "2026-10-05T23:17:02.903",
  "references": [
    {
      "source": "security-advisories@github.com",
      "tags": [
        "Patch"
      ],
      "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
    },
    {
      "source": "security-advisories@github.com",
      "tags": [
        "Issue Tracking",
        "Patch"
      ],
      "url": "https://github.com/vllm-project/vllm/pull/54935"
    },
    {
      "source": "security-advisories@github.com",
      "tags": [
        "Release Notes"
      ],
      "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
    },
    {
      "source": "security-advisories@github.com",
      "tags": [
        "Mitigation",
        "Vendor Advisory"
      ],
      "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
    }
  ],
  "sourceIdentifier": "security-advisories@github.com",
  "vulnStatus": "Analyzed",
  "weaknesses": [
    {
      "description": [
        {
          "lang": "en",
          "value": "CWE-400"
        }
      ],
      "source": "security-advisories@github.com",
      "type": "Secondary"
    }
  ]
}

GHSA-58V5-2M8F-94PR

Vulnerability from github – Published: 2026-10-05 23:42 – Updated: 2026-10-05 23:42

Summary

vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion

Details

Summary

The OpenAI-compatible chat endpoint accepts request-level video loader options through media_io_kwargs. A caller can select the GLMGA sampler and provide large fps and max_frames values. GLMGA first constructs an attacker-sized Python index list and then deduplicates it, even when the supplied video contains only a few frames. This work occurs in the shared media-loading executor before frame reads, allowing a tiny valid video and compact JSON options to consume disproportionate CPU time and memory and delay unrelated media requests.

The demonstrated impact is partial denial of service. No memory corruption, data disclosure, or code execution is claimed.

Affected configuration

The server must expose chat completions for a video-capable model and accept request-level media_io_kwargs. The model does not need to use GLMGA by default because the request value overrides the model's loader mapping. The proof of concept uses an inline two-frame video and explicitly selects the OpenCV decoder, so no external media host or GPU decoder is required.

The request is reachable without vLLM credentials when no API key is configured. With API-key authentication enabled, any caller holding a valid key can reach the same path.

Attack surface

A remote caller submits a valid chat-completion request with a small video and the following request-level options:

{
  "media_io_kwargs": {
    "video": {
      "video_backend": "glmga",
      "backend": "opencv",
      "fps": 500000,
      "max_frames": 500000
    }
  }
}

Increasing both numeric values increases temporary list construction and deduplication work even when the final decoded frame set remains unchanged.

Root cause

ChatCompletionRequest exposes media_io_kwargs as request-controlled nested values: protocol.py#L365-L371.
The request options are carried into chat parameters without a numeric work bound on GLMGA's fps or max_frames: protocol.py#L571-L599.
Media options are merged so request values override configured defaults, including video_backend: connector.py#L577-L605.
The connector selects a registered video loader and runs media loading through the shared executor: video.py#L28-L66 and connector.py#L44-L47.
GLMGA calculates extract_t = min(int(duration * fps), max_frames), builds a list with that many entries, and deduplicates it before decoding frames: video.py#L667-L740.
The OpenCV decoder receives the already-deduplicated indices, so a two-frame file can trigger a large intermediate allocation while producing only two decoded frames: opencv.py#L28-L50.

The missing invariant is a strict upper bound on sampling work before the index list is constructed. Bounding decoded output after deduplication does not bound the vulnerable intermediate computation.

Suggested remediation

Validate request-level video sampling options before dispatching work to the media executor. Enforce conservative absolute limits for fps, max_frames, and especially the computed candidate count. Reject non-finite, negative, or otherwise invalid numeric values.

Avoid constructing O(extract_t) intermediate Python lists. Generate bounded unique frame indices directly from the source frame count and output-frame limit. Apply the limit before list construction and before shared executor submission. Add tests showing that extreme request values are rejected or consume constant memory for a fixed output-frame count.

Workarounds

Remove or filter request-level video_backend, fps, and max_frames options at the gateway.
Do not allow untrusted callers to select GLMGA.
Apply authentication, rate limiting, request concurrency limits, and process memory isolation.
Use a separate constrained media-loading worker pool where operationally feasible.
Existing media byte, pixel, or decoded-frame limits do not necessarily bound this pre-decode candidate-index allocation.

Related advisory

GHSA-cqm8-jxg6-fqfq concerns partial denial of service in the DeepStream video backend through backend confusion and insufficient pixel guarding. This report concerns a different CPU-side GLMGA algorithm: request-controlled sampling values create a large redundant index list before the OpenCV decoder reads a fixed, tiny frame set. The vulnerable backend, resource sink, and required remediation differ.

Severity

5.3 (Medium)


                  
                    CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L

Show details on source website

JSON

To clipboard

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "vllm"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0.23.0rc2"
            },
            {
              "fixed": "0.30.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-105760"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-400"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-10-05T23:42:55Z",
    "nvd_published_at": null,
    "severity": "MODERATE"
  },
  "details": "## Summary\n\nThe OpenAI-compatible chat endpoint accepts request-level video loader options through `media_io_kwargs`. A caller can select the GLMGA sampler and provide large `fps` and `max_frames` values. GLMGA first constructs an attacker-sized Python index list and then deduplicates it, even when the supplied video contains only a few frames. This work occurs in the shared media-loading executor before frame reads, allowing a tiny valid video and compact JSON options to consume disproportionate CPU time and memory and delay unrelated media requests.\n\nThe demonstrated impact is partial denial of service. No memory corruption, data disclosure, or code execution is claimed.\n\n## Affected configuration\n\nThe server must expose chat completions for a video-capable model and accept request-level `media_io_kwargs`. The model does not need to use GLMGA by default because the request value overrides the model\u0027s loader mapping. The proof of concept uses an inline two-frame video and explicitly selects the OpenCV decoder, so no external media host or GPU decoder is required.\n\nThe request is reachable without vLLM credentials when no API key is configured. With API-key authentication enabled, any caller holding a valid key can reach the same path.\n\n## Attack surface\n\nA remote caller submits a valid chat-completion request with a small video and the following request-level options:\n\n```json\n{\n  \"media_io_kwargs\": {\n    \"video\": {\n      \"video_backend\": \"glmga\",\n      \"backend\": \"opencv\",\n      \"fps\": 500000,\n      \"max_frames\": 500000\n    }\n  }\n}\n```\n\nIncreasing both numeric values increases temporary list construction and deduplication work even when the final decoded frame set remains unchanged.\n\n## Root cause\n\n1. `ChatCompletionRequest` exposes `media_io_kwargs` as request-controlled nested values: [`protocol.py#L365-L371`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/openai/chat_completion/protocol.py#L365-L371).\n2. The request options are carried into chat parameters without a numeric work bound on GLMGA\u0027s `fps` or `max_frames`: [`protocol.py#L571-L599`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/openai/chat_completion/protocol.py#L571-L599).\n3. Media options are merged so request values override configured defaults, including `video_backend`: [`connector.py#L577-L605`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/connector.py#L577-L605).\n4. The connector selects a registered video loader and runs media loading through the shared executor: [`video.py#L28-L66`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/video.py#L28-L66) and [`connector.py#L44-L47`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/connector.py#L44-L47).\n5. GLMGA calculates `extract_t = min(int(duration * fps), max_frames)`, builds a list with that many entries, and deduplicates it before decoding frames: [`video.py#L667-L740`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/video.py#L667-L740).\n6. The OpenCV decoder receives the already-deduplicated indices, so a two-frame file can trigger a large intermediate allocation while producing only two decoded frames: [`opencv.py#L28-L50`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/video_decoders/opencv.py#L28-L50).\n\nThe missing invariant is a strict upper bound on sampling work before the index list is constructed. Bounding decoded output after deduplication does not bound the vulnerable intermediate computation.\n\n\n## Suggested remediation\n\nValidate request-level video sampling options before dispatching work to the media executor. Enforce conservative absolute limits for `fps`, `max_frames`, and especially the computed candidate count. Reject non-finite, negative, or otherwise invalid numeric values.\n\nAvoid constructing `O(extract_t)` intermediate Python lists. Generate bounded unique frame indices directly from the source frame count and output-frame limit. Apply the limit before list construction and before shared executor submission. Add tests showing that extreme request values are rejected or consume constant memory for a fixed output-frame count.\n\n## Workarounds\n\n- Remove or filter request-level `video_backend`, `fps`, and `max_frames` options at the gateway.\n- Do not allow untrusted callers to select GLMGA.\n- Apply authentication, rate limiting, request concurrency limits, and process memory isolation.\n- Use a separate constrained media-loading worker pool where operationally feasible.\n- Existing media byte, pixel, or decoded-frame limits do not necessarily bound this pre-decode candidate-index allocation.\n\n## Related advisory\n\n[GHSA-cqm8-jxg6-fqfq](https://github.com/vllm-project/vllm/security/advisories/GHSA-cqm8-jxg6-fqfq) concerns partial denial of service in the DeepStream video backend through backend confusion and insufficient pixel guarding. This report concerns a different CPU-side GLMGA algorithm: request-controlled sampling values create a large redundant index list before the OpenCV decoder reads a fixed, tiny frame set. The vulnerable backend, resource sink, and required remediation differ.",
  "id": "GHSA-58v5-2m8f-94pr",
  "modified": "2026-10-05T23:42:55Z",
  "published": "2026-10-05T23:42:55Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/pull/54935"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/vllm-project/vllm"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L",
      "type": "CVSS_V3"
    }
  ],
  "summary": "vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion"
}

PYSEC-2026-4307

Vulnerability from pysec - Published: 2026-10-15 10:22 - Updated: 2026-10-15 14:38

Details

Summary

The demonstrated impact is partial denial of service. No memory corruption, data disclosure, or code execution is claimed.

Affected configuration

The request is reachable without vLLM credentials when no API key is configured. With API-key authentication enabled, any caller holding a valid key can reach the same path.

Attack surface

A remote caller submits a valid chat-completion request with a small video and the following request-level options:

{
  "media_io_kwargs": {
    "video": {
      "video_backend": "glmga",
      "backend": "opencv",
      "fps": 500000,
      "max_frames": 500000
    }
  }
}

Increasing both numeric values increases temporary list construction and deduplication work even when the final decoded frame set remains unchanged.

Root cause

ChatCompletionRequest exposes media_io_kwargs as request-controlled nested values: protocol.py#L365-L371.
The request options are carried into chat parameters without a numeric work bound on GLMGA's fps or max_frames: protocol.py#L571-L599.
Media options are merged so request values override configured defaults, including video_backend: connector.py#L577-L605.
The connector selects a registered video loader and runs media loading through the shared executor: video.py#L28-L66 and connector.py#L44-L47.
GLMGA calculates extract_t = min(int(duration * fps), max_frames), builds a list with that many entries, and deduplicates it before decoding frames: video.py#L667-L740.
The OpenCV decoder receives the already-deduplicated indices, so a two-frame file can trigger a large intermediate allocation while producing only two decoded frames: opencv.py#L28-L50.

Suggested remediation

Workarounds

Remove or filter request-level video_backend, fps, and max_frames options at the gateway.
Do not allow untrusted callers to select GLMGA.
Apply authentication, rate limiting, request concurrency limits, and process memory isolation.
Use a separate constrained media-loading worker pool where operationally feasible.
Existing media byte, pixel, or decoded-frame limits do not necessarily bound this pre-decode candidate-index allocation.

Related advisory

Severity

5.3 (Medium)


                  
                    CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L

Impacted products

Name	purl
vllm	pkg:pypi/vllm

Aliases

JSON

To clipboard

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "vllm",
        "purl": "pkg:pypi/vllm"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0.23.0rc2"
            },
            {
              "fixed": "0.30.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ],
      "versions": [
        "0.23.0",
        "0.24.0",
        "0.25.0",
        "0.25.1",
        "0.26.0",
        "0.27.0",
        "0.27.1",
        "0.28.0",
        "0.29.0"
      ]
    }
  ],
  "aliases": [
    "CVE-2026-105760",
    "GHSA-58v5-2m8f-94pr"
  ],
  "details": "## Summary\n\nThe OpenAI-compatible chat endpoint accepts request-level video loader options through `media_io_kwargs`. A caller can select the GLMGA sampler and provide large `fps` and `max_frames` values. GLMGA first constructs an attacker-sized Python index list and then deduplicates it, even when the supplied video contains only a few frames. This work occurs in the shared media-loading executor before frame reads, allowing a tiny valid video and compact JSON options to consume disproportionate CPU time and memory and delay unrelated media requests.\n\nThe demonstrated impact is partial denial of service. No memory corruption, data disclosure, or code execution is claimed.\n\n## Affected configuration\n\nThe server must expose chat completions for a video-capable model and accept request-level `media_io_kwargs`. The model does not need to use GLMGA by default because the request value overrides the model\u0027s loader mapping. The proof of concept uses an inline two-frame video and explicitly selects the OpenCV decoder, so no external media host or GPU decoder is required.\n\nThe request is reachable without vLLM credentials when no API key is configured. With API-key authentication enabled, any caller holding a valid key can reach the same path.\n\n## Attack surface\n\nA remote caller submits a valid chat-completion request with a small video and the following request-level options:\n\n```json\n{\n  \"media_io_kwargs\": {\n    \"video\": {\n      \"video_backend\": \"glmga\",\n      \"backend\": \"opencv\",\n      \"fps\": 500000,\n      \"max_frames\": 500000\n    }\n  }\n}\n```\n\nIncreasing both numeric values increases temporary list construction and deduplication work even when the final decoded frame set remains unchanged.\n\n## Root cause\n\n1. `ChatCompletionRequest` exposes `media_io_kwargs` as request-controlled nested values: [`protocol.py#L365-L371`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/openai/chat_completion/protocol.py#L365-L371).\n2. The request options are carried into chat parameters without a numeric work bound on GLMGA\u0027s `fps` or `max_frames`: [`protocol.py#L571-L599`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/openai/chat_completion/protocol.py#L571-L599).\n3. Media options are merged so request values override configured defaults, including `video_backend`: [`connector.py#L577-L605`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/connector.py#L577-L605).\n4. The connector selects a registered video loader and runs media loading through the shared executor: [`video.py#L28-L66`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/video.py#L28-L66) and [`connector.py#L44-L47`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/media/connector.py#L44-L47).\n5. GLMGA calculates `extract_t = min(int(duration * fps), max_frames)`, builds a list with that many entries, and deduplicates it before decoding frames: [`video.py#L667-L740`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/video.py#L667-L740).\n6. The OpenCV decoder receives the already-deduplicated indices, so a two-frame file can trigger a large intermediate allocation while producing only two decoded frames: [`opencv.py#L28-L50`](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/multimodal/video_decoders/opencv.py#L28-L50).\n\nThe missing invariant is a strict upper bound on sampling work before the index list is constructed. Bounding decoded output after deduplication does not bound the vulnerable intermediate computation.\n\n\n## Suggested remediation\n\nValidate request-level video sampling options before dispatching work to the media executor. Enforce conservative absolute limits for `fps`, `max_frames`, and especially the computed candidate count. Reject non-finite, negative, or otherwise invalid numeric values.\n\nAvoid constructing `O(extract_t)` intermediate Python lists. Generate bounded unique frame indices directly from the source frame count and output-frame limit. Apply the limit before list construction and before shared executor submission. Add tests showing that extreme request values are rejected or consume constant memory for a fixed output-frame count.\n\n## Workarounds\n\n- Remove or filter request-level `video_backend`, `fps`, and `max_frames` options at the gateway.\n- Do not allow untrusted callers to select GLMGA.\n- Apply authentication, rate limiting, request concurrency limits, and process memory isolation.\n- Use a separate constrained media-loading worker pool where operationally feasible.\n- Existing media byte, pixel, or decoded-frame limits do not necessarily bound this pre-decode candidate-index allocation.\n\n## Related advisory\n\n[GHSA-cqm8-jxg6-fqfq](https://github.com/vllm-project/vllm/security/advisories/GHSA-cqm8-jxg6-fqfq) concerns partial denial of service in the DeepStream video backend through backend confusion and insufficient pixel guarding. This report concerns a different CPU-side GLMGA algorithm: request-controlled sampling values create a large redundant index list before the OpenCV decoder reads a fixed, tiny frame set. The vulnerable backend, resource sink, and required remediation differ.",
  "id": "PYSEC-2026-4307",
  "modified": "2026-10-15T14:38:12.969063Z",
  "published": "2026-10-15T10:22:45.518163Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-58v5-2m8f-94pr"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/pull/54935"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/commit/8b6de0eb9a09ef53f20cf06bd4d17ee264b9c2a7"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/vllm-project/vllm"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
    },
    {
      "type": "PACKAGE",
      "url": "https://pypi.org/project/vllm"
    },
    {
      "type": "ADVISORY",
      "url": "https://github.com/advisories/GHSA-58v5-2m8f-94pr"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-105760"
    }
  ],
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L",
      "type": "CVSS_V3"
    }
  ],
  "summary": "vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion"
}

Sightings

Author	Source	Type	Date	Other

Nomenclature

Seen: The vulnerability was mentioned, discussed, or observed by the user.
Confirmed: The vulnerability has been validated from an analyst's perspective.
Published Proof of Concept: A public proof of concept is available for this vulnerability.
Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
Not confirmed: The user expressed doubt about the validity of the vulnerability.
Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.

Vulnerability timeline

Sightings and recorded dates, grouped by day · UTC

Order JSON

Publication & reservation Sightings KEV catalogs Other timestamps

Loading timeline…

The MITRE ATT&CK techniques below are AI-generated suggestions, inferred from the description of the vulnerability by the CIRCL/vulnerability-attack-technique-classification-roberta-base model, served locally by ML-Gateway. They have not been verified by an analyst and are provided for guidance only.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.

Related by attack behaviour

Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.

Action not permitted

CVE-2026-105760 (GCVE-0-2026-105760)

FKIE_CVE-2026-105760

GHSA-58V5-2M8F-94PR

Summary

Affected configuration

Attack surface

Root cause

Suggested remediation

Workarounds

Related advisory

PYSEC-2026-4307

Summary

Affected configuration

Attack surface

Root cause

Suggested remediation

Workarounds

Related advisory

Tags

Sightings

Nomenclature

Vulnerability timeline

Related by attack behaviour