SGLang through 0.5.21 contains a resource budget bypass vulnerability that allows remote attackers to override GLM-V video processing limits by supplying fps, max_frames, max_tokens_per_frame and max_image_tokens in the video_url object. Attackers can set these values arbitrarily high so glm4v.py decodes every frame of high-resolution video at full size, exhausting worker memory and denying service.
Fix
Nothing published yet: no patch, vendor advisory or mitigation reference on NVD, no vendor advisory in the News for this CVE, not in CISA's KEV, no fixed package version. BaselineWatch does not guess at one.
Change history 1
- First seenCVSS 8.7NVD modified 2026-10-14 18:17
Affected (CPE criteria) 0
NVD has not attached CPE configurations yet.
References 7
- https://github.com/sgl-project/sglang
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/entrypoints/openai/protocol.py#L592-L594
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/multimodal/processors/glm4v.py#L160-L167
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/multimodal/processors/glm4v.py#L400
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/parser/conversation.py#L691
- https://github.com/sgl-project/sglang/issues/43705
- https://www.vulncheck.com/advisories/sglang-through-0.5.21-resource-budget-bypass-via-glm-v-video-parameters