在 vLLM 0.7.3 至 0.31.0(不含)版本中,客户端提供的 参数会被直接转发给 Transformers 的 方法。当启用 时,这允许客户端选择要导入的模型动态处理器代码的特定修订版本。若攻击者能够控制所服务模型仓库中的某个修订版本,即可向未认证的 端点发送多模态请求,从而在 vLLM API 进程中执行任意 Python 代码。
Although we use advanced large model technology, its output may still contain inaccurate or outdated information.Shenlong tries to ensure data accuracy, but please verify and judge based on the actual situation.
| Vendor | Product | Affected Versions | CPE | Subscribe |
|---|---|---|---|---|
| vllm-project | vllm | 0.7.3 ~ 0.31.0 | - |
|
| # | POC Description | Source Link | Shenlong Link |
|---|
No public POC found.
Login to generate AI POC| CVE-2026-114050 | 7.5 HIGH | vLLM before 0.31.0 Remote Engine Crash via Forged Multimodal UUID |
| CVE-2026-114055 | 7.5 HIGH | vLLM 0.7.3 before 0.31.0 DoS via Unvalidated mm_processor_kwargs |
| CVE-2026-114052 | 7.5 HIGH | vLLM 0.25.0 before 0.31.0 MOSS-Audio Memory Exhaustion via /tokenize mm_processor_kwargs |
| CVE-2026-114051 | 7.5 HIGH | vLLM before 0.31.0 Denial of Service via min_tokens in /inference/v1/generate |
| CVE-2026-114054 | 6.5 MEDIUM | vLLM before 0.31.0 Cross-Request Cache Aliasing via EXIF ImageID and Prefix-Cache Keys |
| CVE-2026-114049 | 3.7 LOW | vLLM before 0.31.0 Stale allowed_token_ids Mask via InputBatch.condense |
No comments yet