---
id: "CVE-2026-115913"
kind: "vulnerability"
aliases: ["CVE-2026-115913"]
title: "vLLM: one long Phi-4 audio clip can exhaust GPU memory and crash"
severity: "medium"
cvss: 6.5
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-tooling", "gpu-inference", "phi-4", "python", "self-hosted", "vllm"]
not_affected_if: ["no-remote-audio-input", "text-only-models"]
published_at: "2026-10-26T17:17:19+00:00"
first_seen_at: "2026-10-26T17:31:22+00:00"
updated_at: "2026-10-26T17:33:17+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-115913", "first_seen_at": "2026-10-26T17:31:22+00:00"}]
---

# vLLM: one long Phi-4 audio clip can exhaust GPU memory and crash

vLLM's Phi-4 multimodal audio encoder uses eager attention by default, so one long audio clip from a remote client makes it store full quadratic attention matrices in GPU memory. The out-of-memory condition terminates EngineCore and interrupts service for all users.

_Summary, check and fix written by a model from the sources; labelled as such._

## How to check

- Check the installed version with `pip show vllm` and treat anything below `0.32.0` as affected.
- Check whether the deployment serves the Phi-4 multimodal model with audio input enabled.

## Fix

- Upgrade vLLM to `0.32.0` or later.

## Timeline

- 2026-10-26T17:31:22+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-115913

## From the sources

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.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_