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.
Affected products
vllm: vllm
References4
- github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6… ExploitPatchVendor Advisory
- github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac98… Patch
- github.com/vllm-project/vllm/pull/60614 Patch
- github.com/vllm-project/vllm/releases/tag/v0.32.0 Release Notes
Metrics
- CVSS 3.1
-
6.5
MEDIUM
Score source: CNA
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H - EPSS — exploit probability
- 0% percentile: 26 · as of 8 November 2026
- CWE
- CWE-770
Could this be exploited in your environment?
A patch closes the known hole. A penetration test shows what an attacker can still reach — with proof.