| Component | vLLM |
| Severity | medium (CVSS 6.5) |
| CVSS vector | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H |
| Published (NVD) | 2026-10-26 |
| Affected versions | < 0.32.0 from the CVE Program record written by the CNA that assigned the CVE |
| Fixed in | 0.32.0 |
| Weakness | CWE-770 |
| CISA KEV | not listed |
NVD description
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.
What to do
- Upgrade vLLM to
0.32.0or later. - Do not expose the component's API or UI to the internet without authentication in front of it.
Sources
- NVD: CVE-2026-115913
- https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928
- https://github.com/vllm-project/vllm/pull/60614
- https://github.com/vllm-project/vllm/releases/tag/v0.32.0
- https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm