CVE-2026-100650

vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.

• CWE-400

• https://github.com/vllm-project/vllm/commit/752a3a504485790a2e8491cacbb35c137339ad34
• https://github.com/vllm-project/vllm/security/advisories/GHSA-p6g9-7v3x-m8mv
• https://www.vulncheck.com/advisories/vllm-before-0.29.0-resource-exhaustion-via-unbounded-media-materialization

Published

Sep 26, 2026

Last Modified

Sep 26, 2026

Status

Received

• How Strix found a critical auth bypass in etcdStrix autonomously discovered a critical authentication bypass in etcd, later designated CVE-2026-33413.
• Autonomous PentestingAI agents that find and validate exploitable vulnerabilities like this one across your applications.
• PR ReviewsPentest every pull request so vulnerable code is caught before it ships to production.
• AI Penetration TestingHow AI-driven penetration testing continuously covers your attack surface.

• CVE-2026-100645SiYuan versions 3.7.0 before 3.8.4 contain a stored cross-si…8
• CVE-2026-100646SiYuan is a self-hosted personal knowledge management system…8.1
• CVE-2026-100647vLLM versions before 0.29.0 contain a denial-of-service vuln…5.3
• CVE-2026-100648vllm before 0.29.0 fails to enforce VLLM_MAX_AUDIO_CLIP_FILE…5.3
• CVE-2026-100649vLLM before 0.29.0 contains a resource-limit bypass vulnerab…3.7
• CVE-2026-10065A weakness has been identified in Shibby Tomato 1.28. This v…8.8
• CVE-2026-100651vLLM before 0.29.0 fails to enforce decoder prompt-length va…6.5
• CVE-2026-100652vLLM versions 0.22.0 through 0.23.0 fail to validate stop_to…5.9
• CVE-2026-100653vLLM is an inference and serving engine for large language m…6.5
• CVE-2026-100654vLLM before 0.29.0 accepts user-controlled stop_token_ids on…6.5
• CVE-2026-100655Netty (io.netty:netty-codec-http) versions up to and includi…7.5
• CVE-2026-100656Netty (io.netty:netty-codec-http) contains an unbounded per-…7.5