CVE-2026-105758
CVE-2026-105758 — vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach
vLLM is an inference and serving engine for large language models. From 0.24.0 until 0.30.0, the Qwen2VLVideoBackend and Qwen3VLVideoBackend classes accept request-level values for the media_io_kwargs.video.max_frames and media_io_kwargs.video.fps fields without enforcing server-side ceilings. An unauthenticated caller can submit these values to the /tokenize endpoint, causing the sampler to decode every frame selected from attacker-controlled video input, consume disproportionate frontend memory, and potentially terminate the API process before scheduling or admission control. The Rust frontend is not affected because it rejects the media_io_kwargs field. This issue is fixed in version 0.30.0.
Published Updated Sources: cvelistV5, GitHub_M
Triage
Is it exploited, how likely is exploitation, what does it touch, and how severe do the scoring sources call it.
Exploitation
Unreported
no source claims exploitation
EPSS
0%
chance of exploitation in 30 days
CVSS base
5.3
MEDIUM
CISA SSVC assessment
Three decision points CISA publishes for the CVEs it assesses · SSVC 2.0.3. A stakeholder decision, not a severity score.
Exploitation
None
none · proof-of-concept · active
Automatable
Yes
can an attacker script all four kill-chain steps
Technical impact
Partial
partial · total control of the vulnerable component
Affected scope
The catalog records vendors and products as separate lists, not pairs, so which product belongs to which vendor is not something this page can say.
Vendors (1)
Products (1)
Every base score collected
Sources score independently and disagree; each row says who scored it and under which version.
| Score | Version | Severity | Expl. | Impact | Source |
|---|---|---|---|---|---|
| 5.3 | CVSS 3.1 | MEDIUM | — | — | cvelistV5 |
Weakness & attack patterns
- CWE-770
Attack patterns reported against this CVE. The ATT&CK techniques below are inferred from its weakness class.
- T1498.001Network Denial of Service: Direct Network Flood
- T1499Endpoint Denial of Service
- T1499.003Endpoint Denial of Service: Application Exhaustion Flood
Detection
Read off the CVSS vector and the weakness class. Starting points, not rules we have tested.
- Search application, proxy, and WAF logs for requests touching /tokenize.
References
4 on the record
- github.com/vllm-project/vllm/security/advisories/GHSA-x6mc-67gf-chw4
x_refsource_CONFIRM
- github.com/vllm-project/vllm/pull/56729
x_refsource_MISC
- github.com/vllm-project/vllm/commit/ea723c81c3ea26425cb69503a5d5e90822a04a45
x_refsource_MISC
- github.com/vllm-project/vllm/releases/tag/v0.30.0
x_refsource_MISC
Elsewhere on this site
- vllm projectevery CVE for this vendor
- CWE-770other pages naming this weakness
Not in any source we poll
Listed rather than left blank: an empty field and an unmeasured one look identical on screen, and only one is a reason to look elsewhere.
- No confirmed IOCs, IP addresses, domains, file hashes, or malware artifacts supplied.
- No organization-specific asset inventory, compensating-control status, or patch deployment evidence supplied.
- No exploit packet captures, log samples, or incident case IDs supplied.