Vulnerability · Published Oct 26, 2026 · Assigned by GitHub_M
CVE-2026-115913vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine
Medium 6.5
In plain English
A medium-severity resource allocation without limits in vllm from vllm-project. There are no confirmed reports of exploitation so far. The chance it is exploited in the next 30 days is estimated at 0.3% (higher than 26% of all known vulnerabilities). Fixed versions are available - upgrade the affected package.
Official 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 packages & fixed versions
| Ecosystem | Package | Fixed in | Advisory |
|---|---|---|---|
| PyPI | vllm | 0.32.0 | GHSA-8397-8qf6-f2gm |
Affected products (as reported)
| Vendor | Product | Versions |
|---|---|---|
| vllm-project | vllm | < 0.32.0 |
Weakness: CWE-770 · Vector: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
References
- https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm x_refsource_CONFIRM
- https://github.com/vllm-project/vllm/pull/60614 x_refsource_MISC
- https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928 x_refsource_MISC
- https://github.com/vllm-project/vllm/releases/tag/v0.32.0 x_refsource_MISC